A Differential Evolution Approach to Optimize Weights of Dynamic Time Warping for Multi-Sensor Based Gesture...

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Machine Learning

Background:

• Gesture recognition systems often struggle with robustness in diverse environments.
• Integrating data from heterogeneous sensors like inertial sensors and depth cameras presents challenges in optimization.

Purpose of the Study:

• To develop a robust, multi-sensory gesture recognition method adaptable to various environments.
• To optimize the fusion of data from heterogeneous sensors using a differential evolution approach.

Main Methods:

• Utilized wearable inertial sensors and depth cameras (Kinect Sensor) for data collection.
• Applied a differential evolution algorithm to optimize dynamic time warping weights for heterogeneous sensor data fusion.
• Evaluated the method on the University of Texas at Dallas Multimodal Human Action Datasets (UTD-MHAD).

Main Results:

• The proposed approach achieved a 10% higher accuracy rate compared to previous methods on the UTD-MHAD dataset.
• Demonstrated improved motion recognition accuracy through optimized sensor data fusion.
• Successfully adjusted the number of depth cameras and combined data with inertial sensors.

Conclusions:

• The differential evolution approach offers a significant improvement in multi-sensory gesture recognition accuracy.
• The developed method provides a more robust solution for gesture recognition in varied environmental conditions.
• Optimizing sensor fusion weights is crucial for enhancing the performance of multi-modal human action recognition.