Multi-target tracking method based on traditional and deep learning algorithms

Abstract

The invention provides a multi-target tracking method based on traditional and deep learning algorithms, and relates to the technical field of computer vision. The multi-target tracking method based on the traditional and deep learning algorithms comprises the steps that a target detector is used for detecting targets in video frames, features of the detected targets are extracted, the apparent features can avoid missing of the targets and the ability to process obstacle targets, and the target tracking accuracy is improved, wherein the motion characteristics mainly depend on streamer, Kalman and other algorithms to carry out target trajectory pre-judgment, then cascade matching and IOU matching are adopted, and then the ID is allocated to the targets. According to the method, the optical flow algorithm with robustness is adopted, so that the phenomenon that the result accuracy is reduced due to the fact that great deviation occurs in Kalman filtering trajectory prediction is avoided. Besides, through the trained yolov5 detection model, real-time verification is carried out through the Deepsort tracking model, and the accuracy of the tracked target is greatly improved.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a multi-target tracking method based on traditional and deep learning algorithms. Background technique [0002] Computer vision has always attracted much attention and has always been a research hotspot. As a part of computer vision, multi-target tracking has also been studied by scholars at home and abroad. Through the study of top journal papers, it is found that most scholars currently use the combination of Kalman filter and Hungarian algorithm to build tracking models. The advantage of this combination is that the detection speed is improved, but the phenomenon of target tracking has not been effectively improved. The present invention effectively avoids the serious deviation of the predicted trajectory of the extended Kalman filter by adding a traditional algorithm-streamer estimation. Contents of the invention [0003] (1) Solved technical problems [0004] Ai...

AI Technical Summary

Problems solved by technology

Through the study of top conference journal papers, it is found that most scholars currently use the combination of Kalman filter and Hungarian algorithm to establish tracking models. The advantage of this combination is that the detection speed is improved, but the phenomenon of target tracking has not been effectively improved.

Images