A template matching tracking method and system based on deep feature fusion

Abstract

The present invention proposes a template matching and tracking method and system based on deep feature fusion, which is realized through the following steps: Step 1, acquire video data, and input the first frame image of the video into a deep convolution network; Step 2, deep convolution The network receives image frame information, performs feature extraction and outputs; step 3, obtains the target position estimated according to the feature information; step 4, judges the target category, and corrects the target frame according to the category information; step 5, extracts the next frame of video data and Input the deep convolutional network for feature extraction; step 6, according to the feature map extracted in step 5, use the target features of the previous frame to perform template matching; step 7, output the target position in the current frame; step 8, judge whether the video is read End, complete target tracking when the end condition is met, and jump to step five if not. The invention realizes target tracking by detecting the target and judging the position of the target in the video.

Description

technical field [0001] The invention relates to a template matching tracking method and system based on deep feature fusion, in particular to the technical field of target tracking. Background technique [0002] With the development of computer technology, detection methods based on deep learning gradually occupy a dominant position in the fields of target detection, classification, and segmentation. Due to factors such as occlusion, illumination, and target non-rigidity in actual scenes, the accuracy and robustness of target tracking still have problems. [0003] In the prior art, when using the depth feature stream to process images, the feature of filtering the still object to establish the feature of the moving object is adopted to propagate the feature of the key frame moving object to the current frame. This technology is prone to drift in the processing process, resulting in insufficient robustness. At the same time, obtaining the depth frame and feature information ...

AI Technical Summary

Problems solved by technology

Due to factors such as occlusion, lighting, and target non-rigidity in actual scenes, there are still problems with the accuracy and robustness of target tracking.

[0003] In the existing technology, when using the depth feature stream to process images, the feature of filtering the static object to establish the feature of the moving object is adopted to propagate the feature of the key frame moving object to the current frame. This technology is prone to drift in the processing process, resulting in insufficient robustness.

At the same time, additional storage costs are required to obtain the depth frame and feature information database, and the feature information sets of all reference contours are calculated and matched one by one, resulting in real-time performance that cannot be satisfied in the actual application process.

Images