Gesture identification method based on deep residual error network

Abstract

The invention discloses a gesture identification method based on a deep residual error network. The gesture identification method based on a deep residual error network includes the steps: 1) acquiring the original data information of gesture, and marking N points to obtain the 2N-dimensional original tag data; 2) preprocessing the original data information and the 2N-dimensional tag data; 3) taking the preprocessed original data information and the 2N-dimensional tag data converted in the hdf5 format as the original training data, inputting the original training data to a deep residual error network to train network parameters, and obtaining a gesture identification model; 4) performing N-point marking being identical to the step 1 on the gesture to be identified to obtain the 2N-dimensional tag data to be identified; and 5) preprocessing the gesture data to be identified and the 2N-dimensional tag data to be identified, inputting the processed gesture data and the 2N-dimensional tag data to be identified converted in the hdf5 format to the gesture identification model for identification, and then obtaining an identification result. The gesture identification method based on a deep residual error network effectively solves the gradient dispersion problem and the network accuracy problem.

Description

technical field [0001] The invention relates to the fields of image processing and object retrieval, in particular to a gesture recognition method based on a deep residual network. Background technique [0002] At present, gesture interaction based on vision has been extensively researched at home and abroad, but due to the ambiguity, space-time differences of the gesture itself, the high latitude, multiple degrees of freedom of the hand, and the incompatibility of the visual problem itself, gesture recognition based on Human interaction platforms for human interaction have always been an inherently challenging research topic. The core point of this project is the recognition of the opponent's shape and the tracking of gestures. There are two main directions in the traditional method: (1) The method based on the hidden Markov chain: GRobel extracts features from the recorded video of the actor wearing colored gloves, and uses the hidden Markov model (Hidden Markov Model, HM...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a gesture recognition method based on a deep residual network, which can effectively solve the problem of gradient dispersion and network accuracy, and at the same time effectively contain the problem of accuracy decline, reducing the training difficulty of the deep network. It greatly improves the accuracy of gesture recognition, and provides a new solution for gesture recognition and even image detection and object recognition. At the same time, this method ensures the accuracy of gesture recognition by inputting multi-dimensional data, and for the input data And the universality of the data format is more powerful, which effectively solves the limitation that the input gesture data can only be one-dimensional

Images