Vision-based obstacle avoidance for unmanned aerial vehicles

A problem of unmanned aerial vehicles (UAVs) being able to avoid obstacles and collisions by means of processing the images taken by one camera is presented in the paper. The optical flow method based on gradient method of Lukas-Kanade was used to obtain the tested structure in 3D space environment (extraction of image's depth). This method allows to generate local information flow patterns from particular neighborhood of surrounding points. Presented algorithm for obstacle location and estimation of its shape was developed and implemented on the workstation with real time uClinux system. Proposed vision-based obstacle avoidance algorithm is demonstrated in simulation and in hardware in the loop flight tests on a fixed-wing UAV.