Human action recognition method based on mars depth feature extraction and enhancement

Abstract

The invention relates to a human action recognition method based on MARS depth feature extraction and enhancement. and the standard cross-entropy loss to train the model; S3: use denoising fusion algorithm to eliminate noise in video data; S4: use image rotation method to simulate camera movement and rotation to simulate human motion in actual real scenes Recognize the scene; S5: Process the human motion through the mosaic occlusion algorithm to recognize the occlusion in the real scene; S6: Use the target scaling method to improve the diversity of the human target size in the real scene, so that the model can continuously learn new data sets; S7: Use The trained and optimized 3D residual transformation model is used for human action recognition. This method is beneficial to improve the accuracy and robustness of human action recognition.

Description

technical field [0001] The invention relates to the field of pattern recognition and computer vision, in particular to a human action recognition method based on MARS depth feature extraction and enhancement. Background technique [0002] In recent years, with the rapid development of computer vision and machine learning, video analysis tasks have shifted from inferring current states to predicting future states. Video-based human action recognition and prediction is such a task, where action recognition is to infer the current state of human actions based on complete action executions, and action prediction is to predict the future state of human actions based on incomplete action executions. These two tasks have become popular research directions due to their explosive emergence in the real world, such as intelligent security video surveillance, human-computer interaction, virtual reality, and medical monitoring. [0003] However, human action recognition based on deep fe...

AI Technical Summary

Problems solved by technology

[0003] However, human action recognition based on deep feature extraction has the following problems: 1) In actual scenarios, the frequency of abnormal actions is very low, and data collection and labeling are difficult, that is, both regular actions and abnormal actions have diverse and complex characteristics. This leads to high diversity within the category and in real scenarios, especially in the field of security, human action recognition based on deep learning has encountered greater challenges

2) In the extraction of deep features, traditional human action feature extraction models cannot accurately and completely extract feature information

In complex scenes, affected by occlusion and camera angle, the traditional feature extraction model is weak, and the robustness of the algorithm needs to be improved.

Images