Target tracking method in multi-camera scene

Abstract

The invention discloses a target tracking method in a multi-camera scene, which realizes splicing of pictures of different cameras by combining YOLO-V4 with an improved DeepSort algorithm and an image splicing algorithm, and finally realizes multi-target tracking in a spliced video. In the aspect of data sets, a self-made intelligent trolley data set and a self-made vehicle re-identification data set containing an intelligent trolley are adopted. According to the vehicle re-identification method, multi-target tracking in a multi-camera scene is realized by constructing a rich data set, improving a model and splicing and fusing pictures, and the vehicle re-identification accuracy is well improved.

Description

technical field [0001] The invention relates to a target tracking method, in particular to a target tracking method in a multi-camera scene. Background technique [0002] Object detection and tracking is a research hotspot in the field of computer vision at present, and has a wide range of applications in video surveillance, automatic driving, human-computer interaction, smart home and other fields. Moving target tracking belongs to the content of video analysis, while video analysis integrates the middle and high-level processing stages in the field of computer vision research, that is, processing image sequences to study the laws of moving targets, or provide semantic and non-physical information for system decision-making alarms. Semantic information support, including motion detection, object classification, object tracking, event detection, etc. As an important branch in the field of computer vision, the research and application of video target tracking methods are inc...

AI Technical Summary

Problems solved by technology

Use the optimized SURF algorithm to match the overlapping parts of the two camera images to complete the multi-camera target handover and achieve cross-camera tracking. Although this method can achieve cross-camera target tracking, it is not intuitive and requires multiple images to be shared. show

Moreover, when there are multiple targets in different camera images, cross-camera tracking can only be achieved for overlapping images, and a definite and accurate id can be assigned to them. However, a suitable solution has not been proposed for how to assign the ids of target tracking for non-overlapping parts. Program

In terms of target detection, the existing technology adopts the frame difference method, that is, the difference operation is performed on two consecutive frames of the video image sequence to obtain the outline of the moving target. This algorithm is simple to implement, low in programming complexity, and fast in operation speed, but this This algorithm relies heavily on the selected inter-frame time interval and segmentation threshold, which has poor versatility and is easily constrained by the scene.

Images