Cross-camera pedestrian re-identification and tracking method

Abstract

The invention discloses a cross-camera pedestrian re-identification and tracking method, which comprises the following steps of: 1) image preprocessing: selecting pedestrian images from a video sequence for pedestrian re-identification, and outputting a clear pedestrian image for pedestrian re-identification by inputting a plurality of low-resolution pedestrian images; 2) cross-camera cooperativeidentification: performing feature extraction of pedestrian images, and performing feature matching of pedestrian images between cameras; 3) blind area track restoration and tracking: a, detecting pedestrians by adopting a target detection method, and inputting images of the pedestrians acquired by a camera in a vision field into a GAN network for restoration so as to realize continuous tracking of a target track; and b, calibrating the image captured by the camera and the real coordinate through a geographic information system to obtain a target starting point and a key position, and integrating through an algorithm to obtain a possible track of the target. According to the invention, cross-camera re-identification and track recovery and tracking are realized, and the efficiency is improved, so that the accuracy is improved.

Description

technical field [0001] The invention belongs to the technical field of human body recognition, and in particular relates to a cross-camera pedestrian re-identification and tracking method. Background technique [0002] Since the end of the 20th century, with the wide-scale application of surveillance camera networks, cross-camera target recognition in an open environment has attracted more and more attention from the academic community as an important technology. Accurate recognition technology can provide powerful help for urban surveillance and security systems. With more and more attention paid to public safety, pedestrian re-identification and cross-camera target tracking have also become popular research areas. [0003] According to reports, in 2018, there were more than 10 million urban surveillance and alarm systems in China, and each person was captured by cameras more than 300 times a day. A large amount of data has potentially huge information, which can be used t...

AI Technical Summary

Problems solved by technology

[0012] 1. Considering the complexity of the open environment, the collected images will be affected by the environment. The first challenge is how to design an algorithm to reduce the impact of the environment on the image

[0013] 2. Considering the variability of pedestrian characteristics, such as adding or removing clothing, opening an umbrella, etc., will affect the efficiency of pedestrian recognition. The second challenge is how to design pedestrian feature extraction and matching algorithms to improve pedestrian recognition efficiency

[0014] 3. Considering the limited coverage of surveillance cameras and the existence of a large number of blind spots, the trajectory of pedestrians cannot be completely monitored. The third challenge is how to design a prediction algorithm to reasonably restore or guess the complete trajectory of pedestrians.

[0015] 4. Considering the large amount of image or video data that needs to be processed, the data collected by a surveillance camera in a city will reach PB level in a day, and the server computing pressure is relatively high. The fourth challenge is how to reasonably design the system structure to reduce the server computing pressure

Images