Research And Application Of Person Re-identification Based On Feature Fusion And Region Segmentation

With the continuous improvement of the application level of urban information technology,intelligent video surveillance system has become one of the key points of urban security work.Because person re-identification technology can quickly find specific pedestrians from massive monitoring data,it has received extensive attention among researchers at home and abroad.However,the complexity of the actual monitoring scene is high,and the variations in pedestrian appearance and posture during the walking process are large,which makes the person re-identification technology face many challenges.This thesis proposes a person re-identification algorithm with feature fusion and region segmentation to handle the changing problem of visual angle and posture.At the same time,the deep features are designed to apply to the actual system.The main research work and innovations of the thesis are as follows:1.Aiming at the large difference in pedestrian appearance caused by the differences in camera angle and person posture,a person re-identification algorithm with region block segmentation and fusion is proposed.Firstly,a segmentation algorithm is designed to capture the more stable local regions in the pedestrian image,and then the information of the key image regions is fully obtained through the fusion of different features.Then a local region removal algorithm is proposed to eliminate the interference block region appearing in the image.The removal algorithm could solve the appearance difference caused by the changes in camera angle and improve the recognition rate.2.A two-task deep learning network is constructed.The loss function of the person re-identification task is combined with the loss function of the attribute recognition task to jointly train the convolution network,which can complete the person re-identification and attribute recognition better.Through the network,pedestrian deep features and attribute features applied to the actual system can be extracted.3.A cross-camera person re-identification and attribute recognition system is built,and the thesis completes the ID and attribute recognition task.Aiming at the problem of gray image caused by insufficient illumination,an image enhancement algorithm is improved to process the poor imaging quality to assist the recognition task.