Action similarity evaluation method based on small sample learning

Abstract

The invention discloses an action similarity evaluation method based on small sample learning. The method comprises the steps of establishing a data preprocessing model, Training model, Test model, ahuman body posture estimation model is adopted to extract a human body overall skeleton motion video and each joint point position; Excluding background interference, the action of the person is splitaccording to the joint point position; setting a sampling pixel value and a sampling interval; intercepting to obtain a sampling video comprising a human body integral skeleton motion video and a joint action taking each joint point as a center; the sampling video combines local information and global information, after data preprocessing, a rewritten triple loss function is used for training, finally video data are mapped to a cosine space, the cosine distance is calculated, and the overall action of a person in the video and the similarity degree result of all joints are output. According to the method, a good action feature mapping model can be learned only by using few samples, so that a good action similarity result is obtained.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to an action similarity evaluation method based on small sample learning. Background technique [0002] At present, the main action similarity evaluation method is to use a dual-stream architecture, namely RGB flow and optical flow. Among them, the RGB stream extracts the spatial features of the people in the video, and the optical flow extracts the motion features of the people in the video, and fuses the dual-stream features. The same operation is performed on both videos to obtain two fused dual-stream features, which are input into the decision network to obtain a similarity score. [0003] The model of the dual-stream architecture generally analyzes the entire picture in general, and pays more attention to the global information. This is caused by the structure of the dual-stream model itself, and the similarity evaluation should pay more attention to the local information. And ...

AI Technical Summary

Problems solved by technology

And because there is a lot of noise in the data, in order to learn good features, the model using the dual-stream architecture needs a lot of data for training to reduce the impact of noise on the final result and alleviate the overfitting phenomenon. In the real environment, the data is relatively A resource that is difficult to obtain, which often requires a lot of cost to obtain data

Images