Flying RoomIdentifying Autonomous Robot Will Bates Chris Newton

Flying, Room-Identifying, Autonomous Robot Will Bates, Chris Newton, Lee Richert, Trip Richert, and Kyle Taylor

Parrot Software License l AR Drone API can be used for “games” » Entertainment » Education » Training l Prohibits » Cartography » Spying » Defense

System Overview l l l Consists of Obstacle Detection, Navigation and Obstacle Avoidance, and Image Processing subsystems Navigation and Obstacle Detection work together to position camera to view room numbers Image Processing done on Raspberry Pi rather than on ground station UDP connection between Pi and AR Drone Sonar Sensors use PWM signal to communicate

Obstacle Detection Uses 5 sonar sensors arranged orthogonally l Sensors read on Pi via PWM signal l Software is interrupt driven l Only 1 interrupt can be cached at a time on the Pi l Testing revealed a scheduling system for the interrupts harms overall performance l

Obstacle Avoidance When obstacle is detected, operation stops and drone hovers l After time period, if object has not moved Drone moves around obstacle and proceeds forward l If movement is detected by sonar sensors, Drone remains in hover state l

Obstacle Avoidance Status Navigation code reviewed, verified for cycle 1 testing and prototype phase l Code modifications being developed for cycle 2 l Integration of Sonars in progress l Once integration is completed modification of avoidance program will begin for cycle 2 full integration l

Ground Station Communication XBee modules are used between the Ground Control Station(GCS) and Robot l Mavlinks protocol replaced with serial communication l Single ASCII character used to represent flight controls and simple messages l

Ground Station Communication Serial connection used to successfully test flight control command code l Serial simplified sending messages to and from the robot l

Computer Vision – Lens Distortion Solution: l Capture imagery with lines (checkerboard pattern) l Model the distortion l Compute reverse mapping Barrel distortion

Computer Vision – Plaque Detection Select informative regions in the image l Numerically describe those regions l Match regions to regions on training images of a room number plaque l For these processes, opencv has open-source code that can be used for these purposes.

Computer Vision – Reading the Plaque Calculate transformation that geometrically aligns training images and plaque in camera image l Extract digits from training images l Use normalized cross-correlation to classify digits at the expected locations on the plaque l

Demo http: //www. youtube. com/watch? v=ZLKo 6 kx TUzs&feature=youtu. be