Vehicle tracking method based on image target detection and proximity analysis

Abstract

The invention belongs to the technical field of image target tracking, and particularly relates to a vehicle tracking method based on image target detection and proximity analysis. The method includes: carrying out target extraction on the video frame by using the target detection technology; obtaining a tracking target candidate area, then calculating a weighted distance between the target of thereference frame and a target in a to-be-tracked image, obtaining a correlation analysis result by combining correlation information of the target of the reference frame and the target in the to-be-tracked image, and finally determining a tracking target result according to the correlation analysis result information. According to the method, the relevance between the target in the to-be-detectedimage and the target of the reference image can be accurately calculated, the high-precision target of the tracking technology can be achieved, meanwhile, multiple frames of images are spliced, targetdetection is completed at a time, the process of target detection for each image is avoided, and the operation rate is high.

Description

technical field [0001] The invention belongs to the technical field of image target tracking, and in particular relates to a vehicle tracking method based on image target detection and proximity analysis. Background technique [0002] In recent years, with the increasing demand for smart parking, the technology related to high-level video smart parking systems is constantly evolving. However, a technical problem that needs to be solved in the current high-level video intelligent parking system is the number plate recognition problem under the condition of license plate occlusion. As a technical method that can estimate the trajectory of the target in the video signal, the target tracking technology can effectively provide the correlation relationship between the target objects between the video frames and the frame. When the license plate is blocked, the target tracking technology can be used to find the vehicle The trajectory of the moving process, and according to the ass...

AI Technical Summary

Problems solved by technology

[0003] 1) Scale adaptive problem in the tracking process: in the process of moving vehicles from near to far or from far to near, the area (scale) of the vehicle imaged in the image is constantly changing, and the traditional correlation filtering method is difficult In the process of re-scale change, the position of the target and the information of the circumscribed rectangular area can be accurately tracked

That is, the traditional method based on correlation filtering can easily lead to problems such as target tracking loss, inaccurate target tracking position, and inaccurate target circumscribed rectangle information.

[0004] 2) The problem of low feature correlation in the tracking process: during the parking process of the intelligent parking system (high-level video intelligent parking system), the inclination of the vehicle body will change drastically, and the vehicle image features will change drastically. The tracking method will be difficult to achieve accurate target tracking under the condition of vehicle tilt angle change (low feature correlation)

[0005] 3) The trade-off between tracking rate and tracking accuracy. The current target tracking method based on correlation filtering has the advantage of fast speed, but the tracking accuracy is extremely poor.

Another target tracking method based on the Siamese network (based on the convolutional neural network) has the characteristics of high precision, but the tracking rate of the tracking algorithm is extremely low without GPU resources. consumption, high precision, high speed requirements

Images