Gesture recognition method based on recurrent 3D convolutional neural network
A neural network and three-dimensional convolution technology, applied in the field of human-computer intelligent interaction, can solve the problems of difficult classification and achieve the effect of low equipment cost and convenient non-contact gesture recognition
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Example Embodiment
[0052] Example one
[0053] Such as figure 1 As shown, step 100 is performed to perform data preprocessing, which is used to process the acquired data into a fixed size to meet the input specification requirements of the input layer of the cyclic three-dimensional convolutional neural network. Such as figure 2 As shown, step 200 and step 210 are executed in sequence to crop the input data, and the randomly cropped size is A×A (in this embodiment, A=112). In step 220, in order to increase the diversity of the training samples, data enhancement is required. Random spatial rotation and scaling are performed on each video, the angle of spatial rotation is ±B° (in this embodiment, B=15), and the zoom size is ±C% (in this embodiment, C=20) . Step 230 is performed to perform random time scaling and jittering on each video, the scaling size is ±D% (in this embodiment, D=20), and the jitter amplitude is ±E frames (in this embodiment, E= 3). Step 240 is performed to obtain data that m...
Example Embodiment
[0059] Example two
[0060] Such as Figure 5 As shown, the overall system architecture consists of four parts: a data input module 500, a data preprocessing module 510, a cyclic three-dimensional convolutional neural network classifier 520, and an output class label 530. The cyclic three-dimensional convolutional neural network classifier 520 can be decomposed into: cyclic three-dimensional convolutional neural network classifier design sub-module 521, cyclic three-dimensional convolutional neural network classifier pre-training sub-module 522, cyclic three-dimensional convolutional neural network classifier training sub-module Module 523, cyclic three-dimensional convolutional neural network classifier optimization sub-module 524 and testing sub-module 525.
[0061] This embodiment proposes a method for gesture recognition based on a cyclic three-dimensional convolutional neural network, which includes importing video data in the data input module 500, performing data preprocessi...
Example Embodiment
[0062] Example three
[0063] Such as Image 6 As shown, step 600 is performed to collect image data through the camera (such as Figure 6a Shown). Step 620 is performed to crop the collected data, remove excess parts, and segment the gesture image (such as Figure 6b Shown). The image is divided into two stages for processing, namely the cyclic 3D convolutional neural network model training stage and the cyclic 3D convolutional neural network model testing stage. In the training phase of the cyclic three-dimensional convolutional neural network model, step 620 is executed to use the data enhancement technique to enhance the data, and the processing result is as follows Figure 6c Shown. Step 630 is executed to preprocess the data and extract clear key frames. Step 640 is performed to train the three-dimensional convolutional nerve first, and then train the overall model. Such as Figure 6d As shown, the method of training is to crop the video to obtain several pictures, and r...
PUM
Abstract
Description
Claims
Application Information
- R&D Engineer
- R&D Manager
- IP Professional
- Industry Leading Data Capabilities
- Powerful AI technology
- Patent DNA Extraction
Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.
© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap