Colour-based object detection and tracking for an autonomous quadrotor

Abstract

Useful applications of Unmanned Air Vehicles (UAVs) include aerial surveillance in hostile military zones or search and rescue operations in disaster stricken areas. The increased visual capacity of UAVs also helps support ground vehicles during scouting missions or to extend communication beyond insurmountable land or water barriers. Computer vision techniques provide a simplistic means to convey information for motion control of a UAV. Hence this work focuses on the development of a vision based image processing algorithm for autonomous navigation of a quadrotor UAV. A camera was used to capture an aerial field of view and transmit a video stream of its perspective to a base station- where OpenCV 2.3.1 vision processing software was used to implement a vision processing algorithm. The algorithm comprises of colour thresholding, the use of image moment and blob detection to detect and track an object within the camea view. Experimental readings of an object's displacement at three altitudes; 1.5m, 2.0m and 2.5m were used to derive pixel-to-cm conversion equations based on the target's pixel coordinates on the viewing window. Through a statistical analysis of variance and standard deviation conducted on 15 experimental readings of displacement ranging from 5cm to 25cm; it was shown that the vision system is best suited for tracking displacements at lower altitude flights. Hence, the best result for variance and standard were achieved when using the derived equation and were 0.64 and 0.8 respectively. The equation derived was also used to derive GPS locking coordinates. Pixel coordinates of a target on the camera display were then used to produce GPS locking coordinates for the quadrotor to track a target object.