A dual-core kcf target tracking method based on spatiotemporal saliency

Abstract

The present invention provides a dual-core KCF target tracking method based on spatiotemporal saliency, comprising the following steps: S1, extracting a target area, using a visual saliency model to extract the salient area; S2, extracting the HOG features of the target frame and the salient frame respectively for training The parameters of the filter; S3, the classification based on the ridge regression classifier calculates the response distribution map y of the filter of the target box and the saliency box respectively k and y s , where max(y k ) and max(y s ) The corresponding coordinate position is the position of the target frame and the position of the salient frame. S4, the offsets of the target frame and the salient frame are calculated based on the position coordinates of the current frame and the previous frame, and the weighted average is obtained to obtain a new offset. The value is used as the correction offset of the target frame; S5, through the position of the previous frame and the correction offset of the target frame, the target position after the correction of the current frame is obtained. The algorithm principle effectively solves the target tracking drift problem.

Description

technical field [0001] The invention relates to the field of target tracking, in particular to a dual-core KCF target tracking method based on spatiotemporal saliency. Background technique [0002] Object tracking is one of the most active research areas in computer vision, and it is widely used in motion analysis, behavior recognition, monitoring, and human-computer interaction. At present, the research of target tracking technology has made great progress, and many tracking algorithms have emerged. The current mainstream tracking algorithms are mainly divided into two categories: [0003] One is the tracking algorithm based on deep learning, which is mainly based on the deep neural network framework for learning. Deep neural networks are used in object tracking due to their powerful learning capabilities in image feature extraction. For example, the target tracking method based on fully convolutional neural network not only uses CNN as a tool for feature extraction, but...

AI Technical Summary

Problems solved by technology

However, the KCF algorithm uses linear interpolation to update the target model, resulting in the accumulation of target apparent information deviations in the tracking process, resulting in target drift, which easily leads to a decrease in target tracking accuracy in subsequent frames

In addition, the KCF algorithm uses a single HOG feature. Although it can capture the outline of the target well, it is easy to cause the target drift tracking to fail when the target is occluded.

Images