Figure automatic tracking method based on edge features and correlation filtering

Abstract

The invention relates to a figure automatic tracking method based on edge features and correlation filtering. The method comprises the following steps: 1), obtaining a current input frame, and extracting a training sample with a current target position as the center; 2) extracting edge features and color features of the training sample to obtain a corresponding feature map, judging whether a filter template needs to be updated or not according to a set rule, if so, executing a step 3), and otherwise, executing a step 4); 3) iteratively updating the filter template by using the feature map of the training sample; 4) respectively carrying out correlation operation on the feature maps corresponding to the edge features and the color features and the filter template, and predicting the targetposition of the next frame; and 5) performing scale prediction on the target position through a scale filter to obtain a target frame of the next frame, and returning to execute the step 2). Comparedwith the prior art, the method has the advantages of real-time performance, accuracy and the like.

Description

technical field [0001] The invention relates to the field of machine vision, in particular to an automatic person tracking method based on edge features and correlation filtering. Background technique [0002] Automatic person tracking based on computer vision Obtaining the image coordinates of the person target from the visual image is the basis for realizing the function of positioning and following the person. The use of a lightweight, fast and accurate automatic person tracking system is of great significance to promote the application of service robots. In addition, this technology also has broad application prospects in areas such as autonomous driving, smart cities, and intelligent monitoring. [0003] Automatic tracking involves the integration of detection and tracking, which belong to two independent research fields. There are two main existing automatic tracking methods: [0004] The first is the method of matching using data association: this method uses data as...

AI Technical Summary

Problems solved by technology

[0004] The first is the method of matching using data association: this method uses data association to achieve optimal matching between the tracking prediction frame and the actually detected observation frame; the existing methods of this type have the following problems: this method is based on detection, and when the target Tracking target switching is prone to occur when occluded or interfered; target feature extraction is performed for each frame as a matching standard, the computational burden is heavy, and it is difficult to meet the real-time requirements

[0008] 1) Generative model method, which tracks based on target modeling, but it does not make full use of background information, relies too much on features so that changes in the appearance of the target itself have a great impact on the results, and needs to process the entire picture, which has poor real-time performance;

[0009] 2) The discriminative model method, which considers the target model and background information at the same time, the method based on manual features is not robust enough in the face of background interference and occlusion, and there is a problem of target drift, while the method based on deep learning uses depth volume The product network extracts the target features, which can improve the target drift to a certain extent, but the real-time performance is poor.

Images