Container commodity tracking method based on optical flow

Abstract

The invention provides a container commodity tracking method based on optical flow. The container commodity tracking method comprises the following steps: S1, training a detector and an optical flow model by using commodity information; s2, object detection is carried out on the video of the container through a detector, and detected commodities are marked on each frame of image of the video through a detection frame; s3, traversing the video frame, and adding the first detection frame into the track as a starting point of the track; s4, for every two continuous frames of images, predicting the position of the commodity marked by the detection frame in the previous frame of image in the next frame of image through an optical flow model, and marking the predicted position as a prediction frame; and S5, calculating the matching degree between the prediction frame and the detection frame in the next frame of image, and adding the detection frame in the next frame of image into the track when the matching degree is higher than a set threshold value. According to the method, the target object is effectively tracked by means of the information of the object optical flow, the problems of missing detection and error detection in the detection process are optimized, and reasoning of the whole model frame in the tracking period is simple and efficient.

Description

technical field [0001] The invention relates to the field of deep learning, in particular to a container commodity tracking method based on optical flow. Background technique [0002] Multi-object tracking (MOT) is a key perception technology required in many computer vision applications, with a wide range of application scenarios, such as autonomous driving and video surveillance. In addition, another important scene is the detection and tracking of multi-target objects based on the container scene. [0003] The most popular multi-target tracking method at this stage is the detector-based tracking method, that is, it needs to track the object based on the detection frame in the image, and track the same detected object between adjacent frames. Due to the problems of occlusion, motion diversity, and less training data of moving objects during motion, tracking methods based on multi-target motion detection are still a challenging task in computer vision. [0004] In the app...

AI Technical Summary

Problems solved by technology

It is not difficult to find that the tracking algorithm that separates feature extraction and data association is inefficient

Images