Accurate Radar Measurements Association with Visual Targets in Traffic Scene

In the field of radar and camera fusion to detect targets, the association between the radar measurements and visual targets is one of the important stages. At present, most of the association methods measure the distance between two points and compare the distance with a fixed threshold to determine whether they are associated or not. However, in complex traffic scenes with multiple types of vehicles, the fixed threshold method cannot reach optimal accuracy of association for each type of vehicles. And the threshold of association box size usually is adjusted based on experience, which also makes it difficult to reach optimal accuracy of association. Aiming at the limitations of existing methods, we propose a method that can adaptively adjust the association threshold based on the visual classification results. And the experiment verifies the improvement of the accuracy of the proposed method compared with the fixed threshold method, especially in the scene of multiple types of vehicles.