Multi-view and Multi-target Association Algorithm for Unmanned Aerial Vehicle Clusters Based on Siamese Network

The multi-target association algorithm of unmanned aerial vehicles (UAV) clusters in unknown environments is used to solve the problem of multiple UAVs, which can correctly determine whether the data from different perspectives contain the same target and eliminate redundant targets. This problem is different from that of multi-target tracking under the same perspective. However, the inherent relationship and differences of data from various perspectives of UAV clusters contribute to the problem of inaccurate target positioning information caused by low-cost detectors and inertial measurement units of UAVs and the constraints of limited communication capability between UAV clusters. Thus, using multi-perspective data is the key to improving the accuracy of multi-perspective situational awareness. This paper proposes a multi-view and multi-target association algorithm for UAV clusters based on the Siamese network. This algorithm can realise the problem of multi-view and multi-target identity judgement with high accuracy, achieve the requirements of high accuracy and high generalisation performance with the depth model based on convolution neural networks, extract the depth features from the optical image and map the original data to the depth feature space. Thus, the same target is matched by comparing the distance measurement of multi-view data in the feature space. The algorithm is verified by an aircraft cluster flight test, which proves the feasibility and accuracy of the algorithm.