Non-overlapping area pedestrian tracking method based on deep neural network

Abstract

The present invention discloses a non-overlapping area pedestrian tracking method based on a deep neural network. The method comprises the following steps of: (1) employing a YOLO algorithm to performdetection of a current pedestrian target in a monitoring video image to segment a pedestrian target image; (2) employing the Kalman algorithm to perform tracking prediction of a detection result; (3)employing a convolutional neural network to extract depth features of images, wherein the images comprise candidate pedestrian images and the target pedestrian images in the step (2), and storing theimages of the candidate pedestrian and features; and (4) calculating similarities of the features of the target features and the features of the candidate pedestrian, and performing sorting of the similarities to identify the target pedestrian. The non-overlapping area pedestrian tracking method can obtain high detection and tracking precision so as to facilitate improvement of pedestrian recognition rate.

Description

technical field [0001] The invention relates to a pedestrian tracking method, in particular to a pedestrian tracking method based on a deep neural network without overlapping areas. Background technique [0002] In recent years, the demand for video surveillance systems has been increasing. For most video surveillance systems, the people appearing in the surveillance video are the focus of attention. Therefore, the intelligent surveillance system needs to have the ability to detect, identify, and track pedestrian targets. ability to further analyze its behavior. Due to the limitations of the camera monitoring area, the joint monitoring of cameras without overlapping fields of view is more and more widely used in monitoring systems, and to realize these functions requires excellent pedestrian detection and pedestrian re-identification technologies. [0003] The patent with the publication number CN 105574515 A discloses "a pedestrian re-identification method under non-overla...

AI Technical Summary

Problems solved by technology

[0004] Purpose of the invention: In order to overcome the deficiencies of the prior art, the present invention provides a method for pedestrian tracking in non-overlapping areas based on deep neural networks, which uses YOLO detection algorithm, Kalman tracking algorithm and deep residual network to detect and track target pedestrians. Tracking, which solves the problem of low detection accuracy and tracking accuracy, and improves the recognition rate of pedestrians

Images