Multi-target vehicle tracking method based on SSD target detection

The invention discloses a multi-target vehicle tracking method based on SSD target detection, and the method comprises the steps: replacing an SSD network structure feature extraction part with a light-weight network, effectively reducing the memory occupation of model parameter operation, and improving the operation speed; selecting four feature response graphs with different resolutions for vehicle detection, carrying out the target position regression and classification on the feature response graphs with different resolutions to obtain category information of a target and position information of a candidate box, and improving the precision of a target detection box; during target tracking, adopting a Hungary algorithm to solve the problem of data association between a prediction resultand a tracking result, and taking the combination of motion similarity and apparent similarity as the total association cost, so the ID Switch phenomenon is effectively reduced, and the accuracy of data association is improv