An autonomous flying object navigated by real-time optical flow and visual target detection

Abstract

We have implemented logic, including real-time optical flow and target detection, using four CMOS video cameras and FPGA mounted on an indoor autonomous flying object. The two techniques presented here, recursive integration and masking optical flow, enhance the signal-to-noise ratio and accuracy of orientation of optical flow. The optical flow of the images from three cameras set to provide a horizontal filed of view can be used to derive the flying object's attitude in terms of roll, pitch, yaw, and height. A target detection process using a camera mounted on top of the flying object can also be used to derive the attitude. These visual processing and nonlinear controlling processes were all implemented in an FPGA (Xilinx XC2V1500). The flying object described here could hover autonomously for several minutes.