Cooperative Target Detection and Recognition for Multiple Flapping-Wing Robots

Abstract

Flapping-wing robots (FWRs) with on-board vision have important applications in military reconnaissance and civil monitoring missions. However, the performance of target detection and recognition for FWRs is partly limited by the usage of a single robot. For such a purpose, this article proposes a three-stage target detection and recognition algorithm to explore the capabilities of multiple FWRs as a cooperative team, which involves target detection, feature extraction, and association matching. First, structure improvement, branch reduction, and knowledge distillation are carried out on the network model of target detection in order to ensure the real-time performance of multichannel image parallel processing. Then, the TransReID algorithm in the field of re-identification and the Hungarian algorithm are used to extract target features and match the target in each camera view, respectively. Finally, experimental results show that compared to the usage of a single FWR, the method proposed in this article can improve the accuracy of target detection and recognition as well as expand the area of cooperative reconnaissance.