An IMU sensor-based personalized gesture recognition system and data processing method

By measuring and removing the gravity vector in the IMU sensor system, implementing multi-layer quality checks and rotation-enhanced training, and combining contrastive learning and knowledge distillation to train a lightweight model, the problems of grip posture differences and insufficient sample size were solved, thereby improving the recognition accuracy and real-time performance of the gesture recognition system.

CN122152138APending Publication Date: 2026-06-05FOSHAN YUNXI CHUANGJING NETWORK TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
FOSHAN YUNXI CHUANGJING NETWORK TECHNOLOGY CO LTD
Filing Date
2026-04-27
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing personalized gesture recognition systems based on IMU sensors suffer from problems such as feature bias due to differences in grip posture, model overfitting due to insufficient user self-sampling, lack of real-time quality control during the acquisition phase, and difficulty in balancing accuracy and inference speed during model deployment.

Method used

By measuring the three-axis gravity vector within a static window and performing degravation processing, a unified preprocessing logic is used to reduce feature bias in the training, inference, and acquisition quality analysis paths. Multi-layer quality checks and rotation-aware data augmentation are implemented, and lightweight models are trained by combining contrastive learning, course learning, and knowledge distillation to improve the model's generalization ability and real-time performance. A dual-gating rejection mechanism is used in the inference stage to reduce the probability of false triggering.

Benefits of technology

It effectively reduces feature bias caused by differences in grip posture, improves the model's generalization ability under small sample conditions, ensures the quality of training data, improves the deployment efficiency and recognition accuracy of the model on resource-constrained terminals, and reduces the probability of false triggering.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122152138A_ABST
    Figure CN122152138A_ABST
Patent Text Reader

Abstract

The application discloses a personalized gesture recognition system based on an IMU sensor and a data processing method, relates to the technical field of human-computer interaction and machine learning, and the system comprises a handheld device and a server, and the handheld device firmware measures and reports a three-axis gravity vector in a static window before collecting a trigger; the server executes gravity removal processing after receiving a data frame, and the same processing logic is shared by three paths of training, reasoning and quality analysis; the gravity-removed data is used for model training and reasoning after unified preprocessing; in the training, rotation perception data enhancement is adopted, the same rotation matrix is applied to six-axis data of the accelerometer and the gyroscope to simulate holding posture changes, and the enhancement amplitude is increased with the training progress; the application can reduce feature deviation introduced by holding posture differences, and is beneficial to improving the generalization ability and deployment efficiency of the gesture recognition model under small sample conditions.
Need to check novelty before this filing date? Find Prior Art