A re-identification algorithm based on multi-information fusion is proposed to solve the problem of perception information redundancy in cooperative vehicle-infrastructure system(CVIS).The inputs of algorithm are the images and trajectories of the two observed targets.By calculating the image similarity and the trajectory similarity,and designing a fusion strategy considering multiple weights,the overall similarity of the two targets can be obtained.Then,a re-identification process is made according to overall similarity of the two targets.For the image similarity,a lightweight image similarity Siamese network is designed;for the trajectory similarity,a trajectory similarity algorithm is designed;for the fusion strategy,four kinds of weights,including image quality weight,trajectory quality weight,image similarity weight,and trajectory similarity weight,are calculated.A fusion function is proposed to realize the fusion of image and trajectory information.A platform of CVIS is built in real traffic scenario.Based on this platform,a variety of experiments under different working conditions are carried out to verify the effectiveness of the algorithm.The results show that compared to the re-identification algorithms based on single information source such as SVDNet,Cam-GAN,MultiScale,the re-identification algorithm based on multi-information fusion improves the re-identification accuracy rate by more than 30％and reduce missed detection rate by more than 20％.
维普
万方数据