Target Detection and Localization for Flapping-Wing Robots Based on Biological Eagle-Eye Vision

Abstract

In recent years, the development of flapping-wing robots (FWRs) has made rapid progress. However, the vision-based perception system of FWRs is still in its infancy, which seriously limits the application of FWRs in military reconnaissance and civil monitoring. To this end, this article designs an eagle-eye-based vision system for target detection, tracking and localization, which fits well with the perception requirements of FWRs. First, inspired by the double-fovea structure of eagle eye, a two-degree-of-freedom pan-tilt equipped with two different focal length cameras is established. Second, the aerial photography characteristics of FWRs and the lateral inhibition mechanism are employed to improve the accuracy of target detection. Third, based on the information of the two different focal length cameras, a cooperative control method is proposed for the pan-tilt to achieve target tracking and the light gradient boosting machine algorithm is adopted to improve the accuracy of target localization. Finally, flight experiments under various scenarios are conducted by using our self-developed FWR named USTB-Hawk, demonstrating the effectiveness of the proposed vision system.