OG-PCL: Efficient Sparse Point Cloud Processing for Human Activity Recognition
Jiuqi Yan, Chendong Xu, Dongyu Liu
TL;DR
This work tackles human activity recognition using sparse mmWave radar point clouds by introducing OG-PCL, a lightweight tri-view CNN-BiLSTM architecture augmented with Occupancy-Gated Convolution (OGConv). The tri-view design preserves spatial cues across three orthogonal projections while maintaining efficiency, and OGConv adaptively gates and compensates features in sparse regions to stabilize responses. Key findings show OG-PCL achieving 91.75% accuracy on RadHAR with only 0.83M parameters, outperforming several baselines and approaching transformer-based methods with far fewer parameters. The approach enables real-time radar-based HAR on edge devices and offers a general sparsity-aware mechanism potentially extendable to other sparse sensing tasks.
Abstract
Human activity recognition (HAR) with millimeter-wave (mmWave) radar offers a privacy-preserving and robust alternative to camera- and wearable-based approaches. In this work, we propose the Occupancy-Gated Parallel-CNN Bi-LSTM (OG-PCL) network to process sparse 3D radar point clouds produced by mmWave sensing. Designed for lightweight deployment, the parameter size of the proposed OG-PCL is only 0.83M and achieves 91.75 accuracy on the RadHAR dataset, outperforming those existing baselines such as 2D CNN, PointNet, and 3D CNN methods. We validate the advantages of the tri-view parallel structure in preserving spatial information across three dimensions while maintaining efficiency through ablation studies. We further introduce the Occupancy-Gated Convolution (OGConv) block and demonstrate the necessity of its occupancy compensation mechanism for handling sparse point clouds. The proposed OG-PCL thus offers a compact yet accurate framework for real-time radar-based HAR on lightweight platforms.