CN-122020272-A - Gesture recognition model building method and device, and gesture recognition method and device

CN122020272ACN 122020272 ACN122020272 ACN 122020272ACN-122020272-A

Abstract

The specification relates to the technical field of artificial intelligence, and provides a gesture recognition model building method and device and a gesture recognition method and device. The method comprises the steps of obtaining inertial measurement data and gesture photocurrent data, determining a first training sample corresponding to the inertial measurement data and a second training sample corresponding to the gesture photocurrent data, inputting the first training sample into a generator of a condition generation countermeasure network model to generate a pseudo sample, inputting the second training sample and the pseudo sample into a discriminator of the condition generation countermeasure network model to obtain a classification result, updating parameters of the condition generation countermeasure network model according to the classification result to obtain a trained condition generation countermeasure network model, and fusing the trained condition generation countermeasure network model and a pre-trained gesture classifier to obtain a gesture recognition model. According to the embodiment of the specification, the deployment cost of human gesture recognition based on photocurrent data can be greatly reduced on the premise of ensuring gesture recognition accuracy.

Inventors

XU WEITAO
WU RUCHENG

Assignees

香港城市大学

Dates

Publication Date: 20260512
Application Date: 20241112

Claims (12)

1. A method for creating a gesture recognition model, the method comprising: Acquiring inertial measurement data and gesture photocurrent data; Determining a first training sample corresponding to the inertial measurement data and a second training sample corresponding to the gesture photocurrent data; Establishing a condition generation countermeasure network model; Inputting the first training sample into a generator of the condition generation countermeasure network model, so that the generator takes the first training sample as a condition and generates a pseudo sample corresponding to the condition by utilizing noise; Inputting the second training sample and the pseudo sample into a discriminator of the condition generation countermeasure network model, so that the discriminator classifies the second training sample and the pseudo sample to obtain a classification result; updating the conditions according to the classification result to generate parameters of an countermeasure network model; repeating the steps of generating a pseudo sample by using the first training sample, judging the pseudo sample and the second training sample, and updating the parameters of the condition generation countermeasure network model according to the classification result until a preset convergence condition is reached, so as to obtain a condition generation countermeasure network model after training is completed; And generating an countermeasure network model by fusing the conditions after training, and obtaining a gesture recognition model by a pre-trained gesture classifier.
2. The method of claim 1, wherein acquiring inertial measurement data and gesture photocurrent data comprises: Acquiring inertial measurement data from the intelligent portable device; And acquiring the gesture photocurrent data from the intelligent wearable device.
3. The method of claim 2, wherein obtaining the gesture photocurrent data from a smart wearable device comprises: When a user interacts with the intelligent wearing equipment, measuring the light intensity and the light incident angle of a light acquisition component in the intelligent wearing equipment and the acceleration data of a target hand of the user and the intelligent wearing equipment in real time, wherein the target hand and the intelligent wearing equipment perform relative movement so as to enable the intelligent wearing equipment to generate gesture photocurrent data; Acquiring initial positions of a target hand and intelligent wearable equipment, shape and size information of the target hand and shape and size information of a light acquisition component; According to the initial position and acceleration data of the target hand and the intelligent wearing equipment, calculating the real-time relative position between the target hand and the intelligent wearing equipment; Calculating an angle threshold between the surface of the light acquisition component and the absorbable light space according to the real-time relative position, the shape and size information of the target hand and the shape and size information of the light acquisition component; and calculating the gesture photocurrent data by using the light intensity, the light incident angle, the angle threshold and the shape and size information of the light acquisition component.
4. The method of claim 1, wherein the generator is constructed using a convolutional neural network, and the arbiter comprises at least two convolutional layers and an activation function layer.
5. The method of claim 1, wherein updating the parameters of the condition generation countermeasure network model based on the classification result comprises: calculating loss by using a preset loss function according to the classification result; Based on the losses, the parameters of the generator and the arbiter in the antagonism network model are generated by updating the conditions with a back propagation algorithm.
6. The method of claim 1, wherein fusing the trained conditions to generate the challenge network model and the pre-trained gesture classifier to obtain the gesture recognition model comprises: and cascading the generator in the training-completed condition generation countermeasure network model with the gesture classifier to obtain a gesture recognition model.
7. A method of gesture recognition, the method comprising: receiving inertial measurement data to be processed; inputting the inertial measurement data to be processed into a gesture recognition model trained by the method of any one of claims 1-6 to obtain a gesture classification result corresponding to the inertial measurement data to be processed.
8. A gesture recognition model creation apparatus, the apparatus comprising: The acquisition module is used for acquiring inertial measurement data and gesture photocurrent data; The determining module is used for determining a first training sample corresponding to the inertial measurement data and a second training sample corresponding to the gesture photocurrent data; the building module is used for building a condition generation countermeasure network model; The generation module is used for inputting the first training sample into a generator of the condition generation countermeasure network model, so that the generator takes the first training sample as a condition and generates a pseudo sample corresponding to the condition by utilizing noise; The judging module is used for inputting the second training sample and the pseudo sample into a discriminator of the condition generation countermeasure network model so that the discriminator classifies the second training sample and the pseudo sample to obtain a classification result; the updating module is used for updating the conditions according to the classification result to generate parameters of the countermeasure network model; The repeating module is used for repeating the steps of generating a pseudo sample by using the first training sample, judging the pseudo sample and the second training sample, updating the parameters of the condition generation countermeasure network model according to the classification result until the preset convergence condition is reached, and obtaining the condition generation countermeasure network model after training is completed; and the fusion module is used for fusing the training-completed condition generation countermeasure network model and the pre-trained gesture classifier to obtain a gesture recognition model.
9. A gesture recognition apparatus, the apparatus comprising: the receiving module is used for receiving the inertial measurement data to be processed; The recognition module is used for inputting the inertial measurement data to be processed into a gesture recognition model trained by the method of any one of claims 1-6 to obtain a gesture classification result corresponding to the inertial measurement data to be processed.
10. A computer device comprising a memory, a processor, and a computer program stored on the memory, characterized in that the computer program, when being executed by the processor, performs the instructions of the method according to any of claims 1-7.
11. A computer storage medium having stored thereon a computer program, which, when executed by a processor of a computer device, performs the instructions of the method according to any of claims 1-7.
12. A computer program product, characterized in that the computer program product comprises a computer program which, when being executed by a processor, executes instructions of the method according to any of claims 1-7.

Description

Gesture recognition model building method and device, and gesture recognition method and device Technical Field The present disclosure relates to the field of artificial intelligence technologies, and in particular, to a method and apparatus for creating a gesture recognition model, and a method and apparatus for recognizing a gesture. Background Gesture recognition based on a solar cell is an emerging technology capable of realizing non-contact and battery-free man-machine interaction, and performs gesture recognition by utilizing photocurrent data generated by the solar cell and different photocurrent modes related to each gesture. Although the technology has great potential in the fields of smart home, automatic driving automobiles, interactive entertainment and the like, the technology still faces great challenges in practical application due to the problems that a large amount of photocurrent data needs to be collected in advance and unfamiliar complicated gestures are difficult to recognize. Therefore, a method for establishing a gesture recognition model is needed to establish and obtain a gesture recognition model, and in an application scene of human gesture recognition through a solar cell, the gesture recognition model does not need to collect photocurrent data from the solar cell, but the sensing data measured through an inertial measurement unit (Inertial Measurement Unit, IMU) of the wearable device realizes cross-mode complex gesture recognition, so that on the premise of ensuring gesture recognition accuracy, the deployment cost of human gesture recognition based on the photocurrent data is greatly reduced. Disclosure of Invention In view of the problem that current solar cell-based gesture recognition requires collection of a large amount of photocurrent data in advance and is difficult to recognize strange complicated gestures, the present solution has been proposed in order to overcome or at least partially solve the above-mentioned problem. In one aspect, some embodiments of the present specification aim to provide a gesture recognition model building method, the method comprising: Acquiring inertial measurement data and gesture photocurrent data; Determining a first training sample corresponding to the inertial measurement data and a second training sample corresponding to the gesture photocurrent data; Establishing a condition generation countermeasure network model; Inputting the first training sample into a generator of the condition generation countermeasure network model, so that the generator takes the first training sample as a condition and generates a pseudo sample corresponding to the condition by utilizing noise; Inputting the second training sample and the pseudo sample into a discriminator of the condition generation countermeasure network model, so that the discriminator classifies the second training sample and the pseudo sample to obtain a classification result; updating the conditions according to the classification result to generate parameters of an countermeasure network model; repeating the steps of generating a pseudo sample by using the first training sample, judging the pseudo sample and the second training sample, and updating the parameters of the condition generation countermeasure network model according to the classification result until a preset convergence condition is reached, so as to obtain a condition generation countermeasure network model after training is completed; And generating an countermeasure network model by fusing the conditions after training, and obtaining a gesture recognition model by a pre-trained gesture classifier. Further, acquiring inertial measurement data and gesture photocurrent data, including: Acquiring inertial measurement data from the intelligent portable device; And acquiring the gesture photocurrent data from the intelligent wearable device. Further, acquiring the gesture photocurrent data from the smart wearable device includes: When a user interacts with the intelligent wearing equipment, measuring the light intensity and the light incident angle of a light acquisition component in the intelligent wearing equipment and the acceleration data of a target hand of the user and the intelligent wearing equipment in real time, wherein the target hand and the intelligent wearing equipment perform relative movement so as to enable the intelligent wearing equipment to generate gesture photocurrent data; Acquiring initial positions of a target hand and intelligent wearable equipment, shape and size information of the target hand and shape and size information of a light acquisition component; According to the initial position and acceleration data of the target hand and the intelligent wearing equipment, calculating the real-time relative position between the target hand and the intelligent wearing equipment; Calculating an angle threshold between the surface of the light acquisition component and the absorbable light space accordi