UAV Use Case: Real-Time Obstacle Avoidance System for Unmanned Aerial Vehicles Based on Stereo Vision

Within the UAV Use Case, we address embedded computer vision application development for unmanned aerial vehicles (UAVs). An example of such an application is obstacle avoidance for UAVs by means of stereo vision. The considered system is based on the computation and analysis of disparity images-images that encode distances to objects in the captured scene. The computed disparity images are used to produce the so-called U-disparity and V-disparity maps that serve as an input to an obstacle detection and avoidance algorithm. While the obstacle avoidance algorithm runs on the ARM processor of the Tulipp platform, the stereo processing algorithm, which consists of an image rectification, a pixel matching, a semi-global matching optimization (SGM), a consistency check, and a median filtering, is run on the FPGA. Although implemented solely with C/C++ and optimized for an accelerated execution on the FPGA fabric with high-level synthesis (HLS), the image processing shows performance suitable for the real-time application on a UAV.