Table tennis target tracking and trajectory predicting method and device, storage medium, and computer equipment

Abstract

The invention relates to a table tennis target tracking and trajectory predicting method and device, a storage medium, and computer equipment. The method includes the following steps: acquiring images of a tracking target taken by two cameras at the same time, and extracting a candidate region corresponding to the tracking target from each image; inputting the candidate regions to a preset tracking model and processing the candidate regions to get a bounding box corresponding to the tracking target; acquiring the 2D coordinates of the center of the bounding box corresponding to the tracking target taken by the two cameras at the same time, and calculating the 3D coordinates of the center of the bounding box according to the projection matrix of the cameras; and acquiring the 3D coordinates of the bounding box at continuous time to constitute a continuous coordinate sequence, inputting the continuous coordinate sequence to a recurrent neural network LSTM and generating a subsequent coordinate sequence through calculation, and getting the trajectory of the tracking target according to the coordinate sequence. The location of the target is tracked using the preset tracking model. Based on the location and LSTM, the advantages of timing features can be analyzed, and the trajectory of the tracking target can be accurately predicted.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a table tennis target tracking and trajectory prediction method, device, storage medium and computer equipment. Background technique [0002] In the design of the table tennis robot system, there are two problems to be solved. One is the problem of object tracking, that is, given the position of the tracked object in the previous frame, predict the possible position of the tracked object in the next frame. The other is the problem of trajectory prediction, that is, given a short sequence of table tennis coordinates, the subsequent coordinate sequence is automatically generated. [0003] As a classic problem in computer vision, object tracking has achieved significant development in recent decades. From the beginning of the Lucas-Kanade tracker, mean-shift tracker, etc. based on pure computer vision methods, to the more complex tracker that integrated detection and its ...

AI Technical Summary

Problems solved by technology

The bounding box obtained by the current tracking algorithm is not accurate enough. The inaccurate bounding box not only means the error of the position information, but also directly causes the entire tracking frame to drift or even lose the target.

When there is an error in target tracking, it will directly lead to an error in trajectory prediction

Images