Collision Avoidance System for UAS Using Stereoscopic Vision
The research experience conducted this summer is about collision avoidance system for unmanned aerial systems (UAS) using stereoscopic vision. It is one of the focus areas of UAS research at the Cal Poly Pomona's UAS Lab. Collision avoidance using stereoscopic vision involves the detection of potential objects that are in the flight path of an aircraft using two cameras or stereo camera, analyzing the trajectory and speed of the hazard, then executing a maneuver that will move the aircraft away from the threat. The algorithm requires the development of a disparity map, which utilizes two cameras to measure the field depth that provides the necessary information required, such as distance from the UAV to the object, and velocity of that object . Two Point Grey Chameleon 3 cameras are mounted on a "Twin-Engine" UAV while using an Intel NUC board for onboard processing. The Intel NUC communicates with the autopilot, Pixhawk, which transmits data to the ground control station via 3DR radios. The Intel NUC generates a disparity map using an algorithm that uses the OpenCV library to process the images into the map. The algorithm will generate the disparity map that will be provided to the collision avoidance algorithm, which will guide the airplane to the location within the map with the least dense area. If one or more of the detected objects are deemed as a collision threat, the avoidance phase is initiated. This calculates the safest path to travel based on the least pixel dense region of the disparity map. The image processing algorithm is designed to remove noise in the image data.