A re-identification method for occluded pedestrians based on centralized learning and deep network learning

Abstract

The invention discloses a re-identification method for occluded pedestrians based on centralized learning and deep network learning. The method uses an occlusion simulator to generate various types of occlusion training samples from original unoccluded training samples, and the generated occlusion training samples are composed of original training samples. The joint training set is used for model training. At the same time, the occlusion and non-occlusion classification losses are added to the pedestrian classification loss, and the multi-task loss function is used instead of the previous single-task loss function to effectively deal with the problem of pedestrian re-identification under occlusion, making the depth When the network learns features, the prior information of occlusion and non-occlusion is considered for feature extraction. Experiments show that the present invention can greatly improve the performance of the existing deep network in re-identification of occluded pedestrians, and has wide application value.

Description

technical field [0001] The present invention relates to a pedestrian re-identification method for occlusion problems, and more specifically, to a occluded pedestrian re-identification method based on centralized learning and deep network learning. Background technique [0002] The task of pedestrian re-identification is to identify the same target object that appears in another camera under one camera. Among them, the occlusion problem is an urgent problem to be solved in pedestrian re-identification. Pedestrian occlusion generally occurs in crowded or complex construction scenes, and these scenes are often accident-prone. For example, a suspect in a dense area may be blocked by pedestrians or other objects such as cars, luggage, and street signs. In this case, the camera captures images of pedestrians with occlusions. We need to search for this complete pedestrian in the pedestrian library or other cameras, which is the task of occluded pedestrian re-identification. There...

AI Technical Summary

Problems solved by technology

[0004] Although the research work on pedestrian re-identification has been well developed, the existing technology almost does not consider occlusion for pedestrian re-identification, and rarely builds models for occluded pedestrian images.

The occlusion part brings interference to the extracted image features, resulting in poor re-identification of pedestrians who encounter occlusion in practical applications

Images