An image gesture action online detection and recognition method based on deep learning

Abstract

The invention discloses an image gesture action online detection and recognition method based on deep learning. The method comprises extracting all frames of images, marking each frame of image, and inputting the marked image into a gesture detection network for training; inputting a section of gesture action video stream into a gesture detection network, obtaining a local gesture area marked as adynamic gesture in each frame of image, and obtaining a local gesture area image frame marked as the dynamic gesture; performing sampling processing by using a piecewise random sampling algorithm, extracting to obtain optical flow information, performing spatial feature map extraction, and inputting the feature maps into respective gesture recognition networks for training; inputting the gesturevideo stream to be detected into a gesture detection network, respectively obtaining predicted classification results, and taking the gesture video stream with the maximum probability as a final result. The gesture actions in the video stream are classified without human intervention, the calculated amount is small, the recognition accuracy is high, the features of the gesture actions can be extracted more efficiently, and the method has higher robustness compared with a complex background.

Description

technical field [0001] The present invention relates to a gesture recognition method, in particular to a method for online detection and recognition of image gestures based on deep learning, which learns highly abstract features through convolutional networks, and has the characteristics of high accuracy and strong robustness. Background technique [0002] In recent years, gesture recognition has mainly stopped at segmenting static gestures in a single simple background, and then using commonly used recognition methods to analyze the meaning of gesture expressions. However, in practical applications, gestures are usually a continuous action in a complex background. How to unsupervised segmentation and recognition of gestures in these complex environments is an urgent problem to be solved. [0003] At present, there are two types of gesture recognition methods based on wearable input device gesture recognition and visual gesture recognition. The wearable input device is easy...

AI Technical Summary

Problems solved by technology

The present invention uses the convolutional network to extract the features of highly abstract gestures in the time dimension and spatial dimension, and automatically combines the features through the convolutional network and maps them to the corresponding gesture categories, which solves the problem that the existing gesture recognition still remains in the right direction. Image recognition and problems requiring human intervention

Images