Phoenix Teams 2015 Unmanned Autonomous System [301664]

Phoenix Team's 2015 Unmanned Autonomous System

Alexandru PANA1, Alexandru Camil MURESAN1, Constantin VISOIU1,

Theodora ANDREESCU2, Petrisor PARVU*,2, Iulian ZAHARACHESCU2,

Claudiu CHERCIU3

*Corresponding author

1INCAS – National Institute for Aerospace Research “Elie Carafoli”,

Technology Development Department,

B-dul Iuliu Maniu 220, Bucharest 061126, Romania

[anonimizat], [anonimizat], [anonimizat]

2“POLITEHNICA” [anonimizat].1-6, RO-011061, Bucharest, Romania

[anonimizat], [anonimizat]*, [anonimizat]

3“POLITEHNICA” [anonimizat], Splaiul Independentei 290, sector 6, Bucharest

[anonimizat]

DOI: 10.13111/2066-8201.2015.7.4.X

Received: 09 November 2015 / Accepted: 20 November 2015

Copyright©2015 Published by INCAS. [anonimizat]-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Abstract: This paper provides a summary of the POLITEHNICA University’s UAS, SkyEye, designed to meet the objectives of AUVSI student: [anonimizat], [anonimizat]. [anonimizat] a gimbaled stabilized point and shoot camera. The transmission of captured images takes place on a 2.4GHz secured wireless link. The received images are then processed for actionable intelligence. [anonimizat] 10 minutes. [anonimizat] a remote 2.4 [anonimizat], or, over a 433 MHz radio link to a [anonimizat] a Remotely Piloted Vehicle (RPV). This paper gives a detailed description of SkyEye’s system.

Key Words: [anonimizat], [anonimizat]

1. INTRODUCTION

This paper provides a summary of the POLITEHNICA University’s UAS, SkyEye, designed to meet the objectives of AUVSI student: [anonimizat], [anonimizat]. [anonimizat] a gimbaled stabilized point and shoot camera.

Fig. 1. The "SkyEye" quad copter

2. DESCRIPTION OF THE SYSTEMS ENGINEERING APPROACH

2.1 Mission requirements analysis

The AUVSI student: [anonimizat]:

[anonimizat] (ISR) support using Unmanned Aircraft Systems (UAS) which has to:

[anonimizat] (SPINS) [anonimizat],

[anonimizat].

be capable of receiving vital messages from Simulated Remote Information Center (SRIC)

provide thermal imaging to locate and track firefighter’s positions.

accurately deliver retardant or water where directed.

Based on the CONOP’s, a Key Performance Parameters (KPP) [anonimizat]:

2.2 Design rationale

The team followed a structured Systems Engineering approach to face the 2015 SUAS challenge. The design process was divided in 3 major phases: 1. Analysis; 2. Preliminary Design; 3. Systems Integration and Testing.

In Analysis phase, the team thoroughly studied the KPP charts prepared to direct the improvements in each sub-system. The team allocated their resources and time on the basis of these KPP charts. The Preliminary Design phase involved subsystems integration, after which extensive laboratory testing and field testing was done. The components were required to perform reliably without any failure. This practice proved beneficial later as it facilitated our systems integration effort, due to dependability of our subsystem modules.

The final phase, Systems Integration and Testing, involved putting together all the subsystem modules on one platform, which were then tested in flight. The behavior and performance of the complete system was observed and necessary alterations were made.

Design rationales have been conducted in the Preliminary Design through Figure of Merit (FoM) tables for key features as follows:

Image capturing subsystem

Image processing subsystem

Aerial platform

Communication system

Ground control

Different solutions were compared against four criteria:

Mission requirements

Safety

Weight

Cost

The imagery system was critical in the design process. A camera with a narrow angle lens was chosen, for more detail on the target. Mounted on a two axis gimbals, the camera is able of being pointed in any direction addressed by the payload operator. All images gathered by the imagery system can be transmitted in real time back to the ground station for evaluation, but most image processing is done onboard the airframe, in background, for time noncritical operations. The aerial platform was designed to be capable of vertical takeoff and landing (VTOL) with the ability to accomplish autonomous flight, including autonomous takeoff and landing, while operating an imagery payload. The use of a rotor-wing aircraft was decided early in the design process. There are several characteristics of a VTOL aircraft that were thought to be beneficial during competition. A feature rotor-wing aircraft boasts over fixed-wing is maneuverability; this is valuable in reconnaissance type missions since the aircraft is able to move three dimensionally at low speeds and low altitudes. Above all other advantages, a VTOL aircraft has the ability to hover. When surveying targets with the intent of gathering actionable intelligence, hovering over an object can supply a much more stable look at a target rather than passing over without stopping.

While building the aircraft, the competition objectives and parameters were addressed during design. The method of autonomy was addressed by implementing a commercially available autopilot. Simple navigation tasks are accomplished by plotting waypoints the aircraft will follow. Re-tasking is as simple as adding additional waypoints for the aircraft to track to. The system was designed to be suitable for surveillance purposes which imply presentation of position and other data transferred during flight. An important goal was the use of commercial off-the-shelf components, making it a low cost option.

2.3 Expected task performance

Performance of the system was streamlined by integrating the imaging system with the autopilot system. The autopilot chosen works seamlessly with the imaging software that powers the camera setup. By integrating these two systems, the issue of complexity by adding extra hardware for camera control and stabilization was eliminated. Advantages that came with this included reduction of weight added by payload, reduced chance of component failure, and overall cost of the system is reduced.

With the VTOL ability of the aircraft, autonomous takeoffs and landings are an unassuming task carried out by the aircraft slowing down, and making a gradual vertical decent onto the landing platform. In the design process of the aircraft, the competition rules and objectives were constantly kept in mind. The rules list a number of parameter thresholds and objectives, and with the ongoing flight test, the aircraft consistently shows that it can operate at the threshold parameters and most of the objective parameters. It is anticipated that the majority of the 40 minutes will have to be used for the mission. The relatively small endurance of the aircraft means at least one change in batteries to complete the mission. Target imagery and location continue to be tested, and with more time spent, it is anticipated that the objective parameters will be met in time for competition. It was found that the area to be scanned is roughly 250000m2 (2.7*106ft2). By field testing of the chosen camera, altitude for reliable target recognition, while keeping the amount of pictures taken reasonably, was found to be 70m (230ft). Under these assumptions, a number of 40 pictures will cover the search area, with a resolution of 2cm/pixel and a ground footprint of 97x72m. The flight time for this survey is 19min. Adding 5min for waypoint navigation before entering the search area and 10min for second path at lower altitude above the identified targets, flight time reaches 34min. This leaves very short time out from the 40min limit, being unsafe to try other secondary tasks. If more than one flight will be scheduled for our team, other secondary task will be attempted, i.e. SRIC task, Emergent target task and Drop test task (for which the payload will have to changed)

2.4 Programmatic risks and mitigation methods

As the rules don’t state the relative importance of the tasks (the scores) it is team guess that most valuable secondary tasks are the ones related to image recognition. The risk is to be wrong with this guess. If this is the case, the team is prepared to revert to manual target recognition, taking only one path over the search area. This will give team the opportunity to try other secondary tasks within the 40 min window.

Also, search area size is to be guessed from little hints given. The risk of wrong guessing the search area size can be mitigating by raising the altitude while searching, for reducing the number of necessary pictures to be taken, and thus reducing the time for covering the entire area to meet the threshold requirements.

The risk of not meeting the design weight of the aircraft will lead to a reduction in endurance which can be mitigated by replacing propulsion battery within the 40 min. window but with the penalty in time for doing this. This will be alleviated by going for manual target recognition and attempting other secondary task after battery replacement.

3. DESCRIPTIONS OF THE UAS DESIGN

3.1 Design descriptions of the aircraft, autopilot system, data link, payloads, ground control station, data processing, and mission planning

Our system is meant to provide real-time recognition of objects on the ground from a UAV, while requiring only a low bandwidth radio link between the aircraft and the ground station. The solution is for the aircraft to do an initial pass of image recognition to find "interesting" objects on the ground, and then to show small thumbnails of those objects on the ground station, overlaid on a satellite map of the area. The operator can then select which of these thumbnails to look at more closely, which will lead to an in-flight mission reconfiguration (new waypoints at a lower altitude).

This will allow bringing down a full high resolution image around that object from the aircraft over the Wi-Fi link. When using this method the operator gets a complete overview of the search area, and can quickly focus on areas of interest. Included in the system is a geo-referencing system that uses the MAVLink telemetry stream along with image timestamps to work out the geographic location (latitude/longitude/altitude/heading) of any region of interest in the image.

3.1.1 Air Vehicle

In order to perform the competition challenge, Phoenix Team has selected a quad copter as an aerial vehicle because of its strong adaptive properties and its ability of carrying extra load. We choose it due to agile maneuverability and advantage of mechanical simplicity. On the other hand it`s capable of hovering and it`s more stable while taking pictures. Due to high resolution pictures for targets extraction, quad copter has good vibration mitigation and strong enough frame. “Sky Eye” is a quad copter made entirely by carbon fiber with arms diameter of 16 mm and its distance between two opposite tubes of 650 mm. The tubes are empty inside for reasons of weight reduction.

The command center, spitted into three main levels, houses the whole system. On the upper level are set the image processing unit, ODROID-XU3, Pixhawk Autopilot, GPS with magnetometer, Wi-Fi, telemetry, power distribution and SSC with engines.

The second level, manufactured at CNC (Computer Numerical Control) by duralumin alloy, supports just the battery packs. On the lower level is fixed the gimbals and QX 10 camera kit. The gimbals is able to point the Sonny QX10 cameradegrees from nadir in the roll and pitch direction. The gimbals is commanded by two mini digital metal gears servo. It is integral manufactured with 0.059 in (1.5 mm) thickness glass textolite. Weights of main parts are shown bellow.

3.1.2 Propulsion System

The propulsion system has several fundamental requirements: provide enough power to complete the mission’s tasks, maximize the cruise speed for area survey all while maintaining a low system weight. The selections of motors were highly important as to the overall performance of the quad copter. The choice was primarily based on the efficiency of the motor, as well as the weight and electrical output of the system. Through the medium of EPROP–calc program, assembly composed of motor-propeller and propulsion pack was computed for 40 min. endurance.

Taking into account current draw, power ratings, size factor and weight a 48-22-490 kV electric motor combined with 16×5.5 carbon propeller performed the challenge due to algorithm sampler. Therefore, 11000 mAh Lithium Polymer cell was chosen to maximize power output while minimizing weight. The cell’s ability to draw high current for a large time interval, as well as maintain a nominally constant output voltage, was an important design constraint. The battery chemistry provides a high-capacity density and a very repeatable charge and discharge cycle and gives a small voltage drop under high amperage loads. Assuming 90% system efficiency it was determined that 1 cell battery configuration was well-suited for the quad copter.

3.1.3 Manufacturing

The manufacturing process included construction of both experimental devices (the main quad copter and its backup). The main parameters considered when deciding upon construction were:

Manufacturability – This includes ease of manufacturing by unqualified personnel

Precision- Precise construction could improve quad copter performance. Creative solutions had to be invented in order to save construction time and maintain high manufacturing precision.

Weight- Ability to manufacture weight efficient parts with enough strength

Commercial availability and price – The commercial availability of parts that the team decided to order like carbon tubes, textolite, and duralumin alloy. The parts delivered had to be bought at a reasonable price and supplied in proper time.

Assembly time – The parts constructed had to be quick to assemble in the field, at the competition.

3.1.4 System Architecture

Fig. 2. System Architecture

The system consists of Pixhawk Autopilot, ODROID-U3, QX10 and GCS. Pixhawk is a high performance autopilot-module suitable for our quad copter, which means it supports both piloted and fully autonomous flight, including GPS waypoints, camera control, auto takeoff and landing. ODROID-U3 is the image processing unit, a powerful credit card size Linux computer with 1.7 GHz Quad-Core processor, 2GByte RAM, 8GB SSD, 100 Mbps Ethernet LAN and High Speed USB2. Image capture is done through a Wi-Fi enabled SONY QX10 camera. The GCS is composed of 2 PC. The first one supports the command, control and monitoring of the UAV flight in real time, providing the flight management of the UAV. The second one receives the processed target images from ODROID-U3, with EXIF positioning data embedded in the images, displays them on a moving map, in real time, and does the feature recognition on them.

Images processing, such as segmentation and recognition, are done with powerful numerical routines from the OpenCV library (Open Source Computer Vision Library) [1]. With the aid of background subtraction, like HSV, mean-shift segmentation and Delaunay filtering, the targets from on-board images are recognized. Crop images of the targets along with EXIF embedded geo-tagging information are transferred to the Ground Control Station for shape recognition and OCR. The recognition algorithm works in two stages, a first stage that finds anything unusual in the image, then a second stage that converts that region to HSV color space and applies some simple heuristics to score the region. User is able to easily plug-in different scoring systems to suit their own image search task.

The cropped images are subjected to numerical search algorithms to extract contours features, which are then classified by their geometric properties (number of corner, convexity of the hull, etc.) in order to recognize their shape and background color. Inner contours are then passed to a neural network system trained to do OCR in order to find the letter in the target. Finally, earth geode routines are used to determine the orientation (heading) of the target.

3.1.4 Autopilot

As a full UAV autopilot, we choose an APM [2], in order to support fully autonomous flight, including hundreds of GPS waypoints, camera control and auto takeoff and landing. APM offers: three axis camera control and stabilization, shutter control, live video link with programmable on-screen-display, data transceivers allow real-time telemetry and control between our ground station computer and APM, including joystick control options, full data logging provides comprehensive post mission analysis, with graphing and Google Earth mapping tools.

3.1.5 Mission Planner (Primary GCS)

Mission Planner [3] is an open-source ground station (GCS) application for MAVlink based autopilots including APM and PX4/Pixhawk that can be run on Windows, Mac OSX, and Linux. Mission Planner allows you to configure a plane, copter, or rover to work with the autopilot to become an autonomous vehicle. Use Mission Planner to calibrate and configure the autopilot, plan and save missions, and view live data in flight.

The above is the main Ground Station view of the Mission Planner, showing the Heads-up Display (HUD). Once you have connected via MAVLink over USB or wireless telemetry the dials and position on this screen will display the telemetry sent by APM.

Fig. 3 Mission Planner

In Mission Planner we can create missions using the easy point-and-click Waypoint Editor. One of the most commonly-used features in pro UAVs is point-and-click mission control in real time. Rather than just pre-planned missions or manually flying the UAV, operators can just click on a map and say “go here now”. Mission Planner is used as primary GCS for controlling autonomous flight, editing waypoints for accessing search area and building waypoints path and commands for area survey. An example is provided in figure below.

Fig. 4. Waypoints editor

3.1.6 MAVProxy (Secondary GCS)

MAVProxy [4] is a fully-functioning GCS for UAV's. The intent is for a minimalist, portable and extendable GCS for any UAV supporting the MAVLink protocol (such as the ArduPilotMega).

It is a command-line, console based app. There are plug-ins included in MAVProxy to provide a basic GUI. For a GUI that works on tablets.

It is written in 100% Python.

It is open source.

It's portable; it should run on any POSIX OS with python, pyserial, and select() function calls, which means Linux, OS X, Windows, and others.

It supports loadable modules, and has modules to support console/s, moving maps, joysticks, antenna trackers, etc.

Due to its modularity, only required MAVProxy modules are used in special designed software for this application. Secondary GCS is used to monitor target recognition, during flight. It uses the map module, flight path module and geo-tagging module. As flight progresses, successive UAV’s positions are dotted on the map and, when a possible target has been found, target thumb is geo-tagged on the map and displayed in a separate window (Fig. 5). The operator has the possibility to accept or reject target for further processing.

Fig. 5. Secondary GCS

3.2 Target types for autonomous detection

3.2.1 Target Recognition and Extraction

Once entered the search area, the UAV follows a predefined path and takes pictures in certain points as to cover all area with 20% overlapping pictures (Fig. 6). Due to altitude limits imposed (min. 100ft., max. 750ft.), picture resolution has to be high (18MP) in order to have enough resolution of the extracted targets, whose dimensions are between 0.6 to 2.4m (2 to 8ft). As seen from Fig. 6 the search area has about 227000m2, and, from 150m altitude, it takes 35 pictures for covering the entire area. Each picture cover a 153x114m rectangle leading to 6cm/pixel resolution. Target resolution will be, under these circumstances, between 10 to 240 pixels. In order to simplify the recognition algorithm, camera stabilization is used, implemented in the Autopilot software. The software assures vertical stabilization of the optical axis. Due to high resolution of pictures, sending them over Wi-Fi to the secondary GCS will take too long. This is why target extraction from the original pictures has to be done onboard UAV, and this is where the ODROID-U3 proves it’s utility. Due to its processing power, it is capable to process one picture and extract targets in less than 10s. Next, only thumbnails of target found are sent over the downlink for further image recognition. Thumbnails transfer time is less than 5s over a 54mB Wi-Fi link. Target recognition and extraction is done by means of background subtraction algorithm. First, the 18MP image acquired by the camera is stored into an array. Each array element contains a pixel’s color information (Red, Green or Blue channels). Next, color space conversion to Hue Saturation Luminance (HSL) of the image is performed. Using histogram back projection algorithm region of interest (targets) in the image are extracted to thumbnails. This algorithm is used for image segmentation or finding objects of interest in an image. In simple words, it creates an image of the same size (but single channel) as that of our input image, where each pixel corresponds to the probability of that pixel belonging to our object. We create a histogram of an image containing our object of interest. The object should fill the image as far as possible for better results. And a color histogram is preferred over grayscale histogram, because color of the object is more better way to define the object than its grayscale intensity. We then “back-project” this histogram over our test image where we need to find the object, i.e., we calculate the probability of every pixel belonging to the ground and show it. The resulting output on proper tresholding gives us the ground alone, which can be subtracted from the image to obtain targets (Fig. 7).

Fig. 6. Search area cover with pictures

Fig. 7. Target (thumbnails) extraction

3.2.2 Finding contours

Contours can be explained simply as a curve joining all the continuous points (along the boundary), having same color or intensity. The contours are a useful tool for shape analysis and object detection and recognition.

Fig. 8. Contours extraction

Image moments assisted us with description of objects after segmentation, estimation of mass center, or object area. Simple properties of the image which are found via image moments include area, total intensity, its centroid and information about its orientation.

The function computes moments, up to the 3rd order, of a vector shape or a rasterized shape. The results are returned in the structure Moments. As an increase of accuracy, Contour Approximation Method helps us to remove all redundant points and compresses the contour. The functions approxPolyDP approximate a curve or a polygon with another curve/ polygon with fewer vertexes so that the distance between them is less or equal to the specified precision. We approximate the basic contour shape to another shape with less number of vertices, by the aid of Douglas-Peucker algorithm [5].

3.2.3 Target Shape Detection

Another problem to solve is the shape of the target. This is done using feature detection algorithms. First, corners of the found contours in the thumbnails are determined. After that, eigenvectors and eigenvalues are found. The output of the function can be used for robust edge or corner detection. The function precornerdetect() calculates the complex spatial derivative-based function of the source image. The corners can be found as local maximums of the function. The function cornersubpix() iterates to find the sub-pixel accurate location of corners or radial saddle points. Sub-pixel accurate corner locator is based on the observation that every vector from the center q to a point p located within a neighborhood of q is orthogonal to the image gradient at p subject to image and measurement noise. The algorithm sets the center of the neighborhood window at this new center q and then iterates until the center stays within a set threshold.

3.2.3.1 Circle detection

The idea of circle detection algorithm is done by first blurring the image, for reducing noise, and then by applying a Hough Circle transformation [5]. Detected circles in the image are sorted out by radius and relevance.

3.2.3.2 Detection of quadrilaterals

Detection of quadrilaterals and triangles is done with help of contour detection functions in OpenCV.

For low resolution images it is essential to filter unwanted corners in the resulting contour, which can be done by simplifying the obtained contour until the number of corners are in the range of commonly used geometrical polygons. The properties of found shapes have to be compared with standard geometrical shapes (triangles, quadrilaterals, other polygons). If the mean quadratic error of the point of found contour from the corresponding points is sufficiently low, we have a "match".

Higher order polygons are difficult to identify, but usually, polygons with more than eight corners are to be considered as being circles.

Fig. 9. Simple shape recognition

3.2.4 Optical Character Recognition

After all this algorithms applied on the contour detection, inner contours are then passed to a neural network system trained to do Optical Character Recognition (OCR). The goal of Optical Character Recognition (OCR) is to classify optical patterns (often contained in a digital image) corresponding to alphanumeric or other characters. The process of OCR involves several steps including: gets the center of mass, find the longest radius, gets the track-step, divide the object into tracks using the track step ,gets the sector step, divide those virtual tracks into equal sectors using the sector step, find relations between adjacent pixels, putting all the feature together. Tesseract [6] open software is used for character recognition.

3.3 Target types supported by autonomous detection (if utilized)

As seen from the detailed design section above the ADLC system does not rely on templates in recognizing targets characteristics, except OCR. Due to this, team is confident that the target recognition algorithm will be capable of treating autonomously any type of target within the limits stated by the rules (2 to 8ft, basic geometric shape, 50 to 90% alphanumeric character). Yet, one possible problem might be the description of target and character colors. While system can find mean color in numeric BGR format, a standard literal classification of colors is not available. Although, team has found over 800 colors names over the internet, the possibilities being much more (1.6billion colors), we have to rely on “nearest best match” to state the color, which can lead to mistakes, when done automatically.

3.4 Mission tasks being attempted

As the SUAS committee’s intent to supply more tasks than can be completed in the available mission time, and as team has found, analyzing KPP, mission tasks being attempted have to be prioritized. Primary tasks are of paramount importance in achieving a good score. So they are mandatory tasks from team’s point of view. The whole design is focused in fulfilling the requirements of these tasks: AUTONOMOUS FLIGHT TASK and SEARCH AREA TASK. As stated before, depending on the evolution of the contest, secondary tasks will be attempted in a priority order chosen to increase the team score, without jeopardizing the fulfillment of primary tasks. ADLC and ACTIONABLE INTELLIGENCE tasks will be attempted if the assumptions in §1.4 are correct. If not, and attempting these task will lead to the risk of going over endurance of the aircraft, these task will not be attempted. IR SEARCH task will not be attempted because a suitable sensor was not available to our team due to financial restrictions. OFF‐AXIS TARGET, EMERGENT TARGET, and AIR‐DROP task will only be attempted in the event flight time will be available to team, because they require either software or hardware reconfiguration of the aircraft (video streaming, imagery payload replacement with drop device). SRIC, INTEROPERABILITY and SDA tasks will only be attempted if the primary tasks are fulfilled and enough mission time is left for safely undergoing this task, because it does not require reconfiguration of the air vehicle, except Wi-Fi network software configuration which can be easily done through simple command line connection to the on-board ODROID computer.

4. TEST AND EVALUATION RESULTS

4.1 Mission task performance

4.1.1 Flight testing

As of May 24th, a total of 345 minutes of flight including 26 autonomous takeoffs and landing cycles were accomplished by the team. Throughout the flight testing “SkyEye” has shown the ability to take off in winds up to 15m/s (30kts), maintain a forward velocity of 5 to 20m/s (10 to 39kts), and demonstrated a programmable climb and a descent rate between 0.5 to 5m/s (1 to 10kts). To complete autonomous take off and landings a Sonar sensor was installed to assist the autopilot, from the first test flight autonomous takeoffs and landing have been carried out with less than 5 feet of error from the designated landing area. When testing the waypoint tracking feature of the APM, a qualified external pilot is on standby in case the aircraft performs undesirably. During test flights the aircraft is commanded through the GCS to fly to assigned waypoints, and consistently carries out the function with no measurable error.

4.1.2 Endurance

“SkyEye” has a usable flight battery capacity of 10,000mAh. The first 50% of the battery amperage is equal to 70% of the total flight time. The last 50% of usable battery capacity is equal to 30% of the total allowable flight time. This allows approximately 34 minutes of endurance. Test have shown that the target endurance could not be meet with the available (used) batteries in a safely manner. During one of the flight tests, after a long wait time with system idle on the ground, the batteries reached the failsafe threshold (13V) after only 24min. in the air, which caused the aircraft to land prematurely, before the mission end. This will be corrected by changing the propulsion batteries with fresh new ones, if financial restrictions will allow it, or by reducing idle time; propulsion batteries will have to be connected only a short time before takeoff.

4.2 Payload system performance

4.2.1 Payload Testing

The payload was tested during initial RC flight testing. The weight of the payload was kept at minimum to achieve the desired endurance. Utilizing a qualified RC pilot weight was added to the aircraft until the lower bound of flight time (34min) was reached. The copter was capable of lifting 2lbs, with a MTOW of 6lbs. Next, gimbals stabilization and vibration dumping was tested, by streaming the images over Wi-Fi while flying at different speeds and orientations. Max. pitch and roll angles for steady shot were found to be 32o, allowing horizontal speeds up to 10m/s (19knots).

4.2.2 Imagery System Testing

The imagery testing began during the final phase of our flight testing. Minor adjustments continue to be made after each test flight to improve image quality. Vibration of the camera is an issue that is constantly addressed and changes to the gimbals have reduced the problem to manageable levels. Operation of the Imagery payload is practiced during every test flight to insure the highest experience level in time for competition.

4.2.3 Target Location Accuracy

According the rule book’s key performance parameters the objective is to find the target within 50 feet. To meet this objective the team took target data from practice mission and compared this data to the actual location of the target. The first tests of our target location showed that the location of the majority of the targets were 50ft-100ft from the actual position of the target. After inspection, the team found the camera mount had been 4o lower than the airframes level position. In order to fix this the gimbals calibration setting needed to be adjusted in GCS.

4.3 Autopilot system performance

Flight tests have shown a very good guidance performance in autopilot mode with 2m (less than 7ft) radius on waypoints as can be seen from analyzing telemetry logs of flown missions (Fig. 10). Loiter mode proved very satisfactory in winds bellow 5m/s(1kts), and acceptable in stronger winds up to 15m/s (30kts), with a maximum 5m(15ft) position deviation from the programmed waypoint. Altitude hold is proved to be accurate, within a 1m(3ft) margin. All Failsafe modes (RC Failsafe, GCS Failsafe, battery Failsafe) proved to work as expected. During test flights done RC Failsafe and GCS Failsafe did not happened in normal operation, but only in forced test mode (RC transmitter shut down intentionally or telemetry link cut on purpose).

4.4 Evaluation results supporting evidence of likely mission accomplishment

After evaluation of the test results presented above and carefully analyzes of many flight logs, team is confident that SkyEye is prepared for accomplishing selected missions. Proof of primary missions’ accomplishments can be seen in flight video provided.

Fig. 10. Mission telemetry log

5. SAFETY CONSIDERATIONS/APPROACH

In the UPB-FIA UAS department, safety is applied intrinsically. By the use of checklists and redundancy in its systems safety is constantly monitored and addressed. This type of safety carried over to team operations, and is applied to all components of our missions. For this competition, safety was addressed in three parts: Aircraft, mission and redundancy.

5.1 Specific safety criteria for both operations and design

5.1.1 Aircraft

During the assembly phase of the SkyEye copter, close attention was paid to insure the aircraft was properly assembled as per design. The torque specifications were checked and Loctite was used where applicable. Also on the aircraft, special attention was paid to making sure all loose cables and components are properly secured to reduce the risk of unintended movement during flight operations.

Additional safety precautions that were taken on the aircraft included high visibility paint. Batteries are wrapped in a bright blue to promote high visibility to both operators and onlookers. Before integrating the autopilot system, several external pilot test flights were conducted on the aircraft to insure functionality of the copter, along with payload and endurance tests mentioned in the design portion. Safety of the operators and the surrounding environment is a prime concern in an autonomous mission. In the event of any equipment failure, the autopilot failsafe is realized in three ways. The first is the RC failsafe that continuously keeps a track of RC radio link, losing which it engages failsafe RTL. The second is a PPM multiplexer which allows the RC pilot to manually over-ride the aircraft any time during flight. The third is the telemetry link that enables GCS operator to take control, transforming the aircraft in a RPV. Safety is a major concern during any UAS operation. To ensure hassle-free mission execution, every flight crew member has a designated check-list which includes reaction under emergency situations. During flight operations under normal conditions, the RC pilot receives commands only via the Flight Director. While the aircraft is in air, the Safety Officer is stationed near the pilot. In case of emergency landing announcement from the pilot, he executes operational procedures to ensure the landing spot is clear and the overall safety of spectators and flight crew.

5.1.2 Mission

A risk assessment tool designed by our safety officer is used before every flight. After completion, this tool gives the operation a go/no go decision based on mission type, environmental factors, and crew readiness. The risk assessment tool relies on a point system where low numbers represent small risk, and larger numbers represent high risks. When added together these points help decide a go/no go decision.

Likewise every mission performed by the SkyEye copter always begins and ends with a checklist. A point is made to never commit anything to memory so that no item is ever missed. Before the flight portion of the mission begins, a crew briefing is held, here the crew assignments are verbalized along with type of takeoff, mission objectives, emergency procedures, and recovery methods. In the event of an emergency or unintended movement, the crew is instructed to take shelter inside, or behind the GCS to prevent being struck by the aircraft. Crew assignments are a very important part of mission safety. It is imperative that every person participating in a flight test is given an assignment to keep crew focused on the task at hand, and know what their role is in case of an emergency. The ground control station consists of the parts of the crew who conduct the autopilot and payload operations. It is their job to conduct the checklist before the flight, and commanding operations during flight. The ground station crew is responsible for all operations of the aircraft, unless operations are transferred the external pilot. In the event of an emergency, the GCS crew is required to stay inside the station to prevent injury. Ground crew members are responsible for the handling of the aircraft when not flying. They are tasked to transport the aircraft, charge batteries and test systems before each flight. Once all checklists are complete, the ground crew must move to inside or behind the ground control station during flight operations. The safety pilot’s responsibility is to handle the external pilot transmitter at all times. When necessary the safety pilot assumes control of the aircraft, and is ultimately responsible for the safety of flight operations. While in flight, the safety pilot operates as a spotter to insure clearance from danger. The team has established a go/no-go criteria based on previous experiences which are strictly adhered to. SkyEye shall not fly under the following circumstances:

If there is any precipitation.

If there is an approaching thunder-storm.

If the visibility is less than 1 mile.

If GPS lock fails.

If there is any perceptible damage during ground operations.

If range test fails before 200 feet on the primary RC link with all wireless devices operational.

If winds exceed 20 knots.

If there is low light.

5.2 Safety risks and mitigation methods

Based on past failures and possible risks experienced from several flight tests, a detailed Failure Modes and Effects Analysis (FMEA) approach was employed to devise a meticulous Risk Mitigation Protocol to be followed by flight crew during flight tests. This practice ensures the system’s flight readiness to meet the competitive safety requirements of AUVSI SUAS 2015.

5.2.1 Redundancy

Redundancy begins in our ground control station (GCS). The UPB-FIA GCS boasts a triple redundant power system. Since UAS operations cannot be carried out without the GCS, these redundancies are very important. The entire electrical system is powered by external supply power provided. Once that fails the system relies solely on the laptop batteries that also last approximately 90 minutes. So the GCS is capable of operating without the generator for approximately 90 minutes, which is more than enough time considering the flight time of our aircraft is only 34 minutes.

The APM autopilot system also boasts redundancy. During any phase of flight the abort function can be initiated. Depending on what phase the aircraft is in, this feature can safeguard the aircraft and its surroundings. If abort is selected during rotor spin-up or liftoff the rotors will spin down and the aircraft will kill the motor and put itself into prelaunch mode. At any time during flight operations abort is selected, the aircraft will decelerate to zero airspeed, and begin a controlled decent and land directly below its flight path. Also if battery failsafe condition occurs the same landing strategy will be engaged. This is a great benefit of having an aircraft capable of vertical takeoff and landing (VTOL). Where a fixed-wing aircraft needs a large area for emergency procedures, a VTOL aircraft can simply land in almost any area with little risk to damaging the aircraft or onlookers. If at any moment communication is lost, the RTL plan programmed into the aircraft for lost communication goes into effect. Before the mission, the operator chooses what the lost communication procedure will be. In our operations, lost communications always results in the aircraft initiating its RTL flight plan, and executing an autonomous landing when arrived at home location. This eliminates any possibility of loss of aircraft due to communication error.

5. CONCLUSION

Our quad copter equipped with the systems described earlier performed through AUVSI 2015 Competition. In the middle of the race, we accomplished a fully autonomous flight in order to capture and process specific targets existing on flight path. After 4 days of competition, we won 4 Place Overall and special prize for “The best rotary wing”.

Fig.11. Best rotary wing design prize

Fig. 12. 4th Place Overall – prize

REFERENCES

[1] OpenCV with Python – Prateek Joshi

[2] ArduCopter wiki – http://copter.ardupilot.com/

[3] Mission Planner wiki – http://planner.ardupilot.com/

[4] MAVProxy wiki – http://dronecode.github.io/MAVProxy/html/index.html

[5] Learning Image Processing with OpenCV – Gloria Buenco, Oscar Deniz Suarez, Jose Luis Espinosa Aranda, Jesus Salido Tercero, Ismael Serrano Gracia, Noelia Vallez Enano

[6] Tesseract-OCR wiki – https://code.google.com/p/tesseract-ocr/