Abstract:

Because of the high field of view angle and wide detection range， roadside sensor is conducive to observe the spatial?temporal relationship and movement trend of targets， and becomes an important sensing means to support the vehicle road cloud integration system. However， the current roadside sensor faces the challenge that it is difficult to continuously track the target motion across the sensing domain. In order to solve the problem of target handover between cross view cameras， a bi?directional matching target handover method is proposed for roadside sensing overlapped field of vision. The transformation relationship between cameras is established to project the vehicle target in the overlapped domain， and the matching constraint mechanism is designed to solve the problem of matching conflict. The bidirectional matching strategy is designed based on the bidirectional projection of the target， and the handover result is output according to the matching matrix identifier and color similarity. The roadside experimental data show that the proposed method can be effectively applied to typical urban road traffic scenes. Compared with the one?way matching method， it reduces the missing matching by 22.22% and the false matching by 28.57% for road scenes； The missing matching and false matching of intersection scenes are reduced by 28.89% and 44.00%， respectively， which has good accuracy and reliability.

Key words: vehicle?road?cloud integration, roadside camera, overlapping field of view, target handover, bidirectional matching