Fixed Point Cloud Normalization and None-Sequential Modeling for Hand Gesture Recognition Based on Short-Range mmWave Radar Sensor’s Sparse Time-Series Point Cloud

Abstract

This article introduces a novel approach to hand gesture recognition utilizing sparse time-series point cloud data obtained through a short-range mmWave radar sensor. Our proposed method not only mitigates the need for complex data format conversions but also operates efficiently with sparse time-series point cloud data, leading to significant advantages in processing time and storage consuming. This study focuses on accurately classifying point cloud sequences representing hand gestures by none-complex sequence modeling. The proposed methods include a modified PointNet configuration suitable for gesture recognition and an optimized point cloud data preprocessing. The sequential features of input data applied to the proposed model by integrating frame order information into the vector representation of each point and using point augmentation and sampling to normalize the point cloud that is measured differently depending on the type of hand gesture and position. The performance of a point cloud-based recognition model with a sparse matrix form can be improved by ensuring the preservation of a fixed input shape. Performance experiments demonstrate the superiority of the proposed methods in classification performance compared to the existing methods in the recurrent neural network (RNN) series and PointNet. The experimental results provide insights for selecting optimal parameters in specific application environments. In conclusion, this study presents a robust system for hand gesture recognition, offering accurate classification of point cloud sequences without the need for complex data format conversion. The simplicity of data processing and reduced computational cost are notable advantages, contributing to the development of cost-effective and efficient hand gesture recognition systems.