Enhanced Sparse Point Cloud Data Processing for Privacy-aware Human Action Recognition
Maimunatu Tunau, Vincent Gbouna Zakka, Zhuangzhuang Dai · 2025
Read Enhanced Sparse Point Cloud Data Processing for Privacy-aware Human Action Recognition by Maimunatu Tunau, Vincent Gbouna Zakka, Zhuangzhuang Dai (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-10469.txt.
Introduction
Cameras in a home can feel like spies, so a wave based sense that reads moves as sparse dots is a soft path to watch acts. Maimunatu Tunau, Vincent Gbouna Zakka, and Zhuangzhuang Dai tune that path in a 2025 study. The dots come thin and loud, so clean steps must run first.
Three known steps get a full test: each alone, in pairs, and all three at once. One is a clump rule that drops stray dots. A match rule pairs dots frame to frame. A smooth track rule then evens the path. The team also boosts each step on its own. Marks from a known set weigh truth gain next to time cost. The map of gains helps crews pick a mix that fits their purse.
Catalog record
| CREATOR | Maimunatu Tunau, Vincent Gbouna Zakka, Zhuangzhuang Dai |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 23 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-10469
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Enhanced Sparse Point Cloud Data Processing for Privacy-aware Human Action Recognition Maimunatu Tunau, Vincent Gbouna Zakka, Zhuangzhuang Dai Abstract Human Action Recognition (HAR) plays a crucial role in healthcare, fitness tracking, and ambient assisted living technologies. While traditional vision-based HAR systems are effective, they pose privacy concerns. mmWave radar sensors offer a privacy-preserving alternative but present challenges due to the sparse and noisy nature of their point cloud data. In the literature, three primary data processing methods—Density-Based Spatial Clustering of Applications with Noise (DBSCAN), the Hungarian Algorithm, and Kalman Filtering—have been widely used to improve the quality and continuity of radar data. However, a comprehensive evaluation of these methods, both individually and in combination, remains lacking. This paper addresses that gap by conducting a detailed performance analysis of the three methods using the MiliPoint dataset. We evaluate each method individually, all possible pairwise combinations, and the combination of all three, assessing both recognition accuracy and computational cost. Furthermore, we propose targeted enhancements to the individual methods aimed at improving accuracy. Our results provide crucial insights into the strengths and trade-offs of each method and their integrations, guiding future work on mmWave-based HAR systems.Our source code is made publicly available at 1 1 1 https://github.com/Maimunatunau/Human-Action-Recognition-HAR-using-mmWave-Radar . 1 Introduction In the era of intelligent environments and pervasive computing, the ability for systems to perceive and respond to human actions in real time has become fundamental to the development of smart technologies. From adaptive lighting and automated security in smart homes to fall detection systems for elderly care, Human Action Recognition (HAR) plays a crucial role in enabling responsive, context-aware applications. Traditionally, HAR has relied on camera-based systems due to their rich visual information. However, these systems come with some limitations, including privacy concerns, dependence on good lighting conditions, and high computational demands [ 17 ] . Millimeter-wave (mmWave) radar has emerged as a promising alternative for HAR, offering privacy-preserving, lighting-invariant sensing capabilities. Operating in high-frequency bands, mmWave radar enables robust motion detection even in visually obscured environments [ 16 ] . When combined with deep learning models, this sensing modality has achieved considerable success in recognizing human actions [ 9 ] . Still, challenges remain—particularly in dealing with the sparse, noisy point cloud data generated by radar sensors, which can hinder both feature extraction and accurate classification [ 12 ] . To address these challenges, several preprocessing approaches have been introduced in recent studies, focusing on clustering, tracking, and temporal association. For example, [ 18 ] introduced the mID pipeline, combining DBSCAN clustering, the Hungarian Algorithm, and a Kalman Filter, but their work lacked systematic tuning and degraded with sparse input. Other studies explored deep learning enhancements to Kalman filtering [ 19 ] , and [ 20 ] emphasized fast clustering and association strategies. While each of these approaches contributes valuable insights, they are often evaluated in isolation. A unified and systematic evaluation of these methods—and their combinations—is yet to be explored. This paper addresses that gap by presenting a comprehensive evaluation of three widely used radar point cloud preprocessing techniques: DBSCAN, the Hungarian Algorithm, and Kalman Filtering. Our focus is on analyzing how each method performs individually, how they interact in pairs, and how they behave when combined—all using the MiliPoint dataset as a case study. We assess both recognition accuracy and computational efficiency to better understand the trade-offs involved in using each method or combination. In addition, we propose practical enhancements to the individual techniques to boost performance, particularly in noisy environments. By conducting this systematic evaluation, we aim to inform the design of more effective mmWave radar-based HAR systems and provide guidance for selecting or combining the preprocessing methods. Figure 1: Challenges and opportunities in mmWave radar-based Human Action Recognition. Left: Raw mmWave data is noisy, sparse, and lacks meaningful structure. Middle: Integration and application of data processing pipeline. Right: Our proposed pipeline (DBSCAN, Hungarian Algorithm, Kalman Filter) mitigates noise, preserves temporal coherence, and enables reliable action recognition. 2 Related Work 2.1 Evolution of Human Action Recognition (HAR) HAR has undergone substantial development over the past two decades, driven by advancements in artificial intelligence, sensor technologies, and computational capabilities. Early HAR systems predominantly relied on vision-based techniques such as Optical Flow and Histogram of Oriented Gradients (HOG), which extracted motion and shape descriptors from video sequences [ 4 ] . While these approaches achieved moderate success in controlled environments, they were highly sensitive to changes in lighting, occlusions, and raised concerns regarding user privacy. The introduction of machine learning algorithms, for instance Support Vector Machines (SVMs), marked a step forward in HAR by enabling supervised classification of complex motion patterns [ 14 ] . However, these approaches required extensive feature engineering, and often struggled with scalability in multi-class action recognition settings. More recently, deep learning particularly Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) has largely replaced the classical models by automating feature extraction and improving recognition performance across diverse datasets [ 9 , 10 ] . Despite these improvements, vision-based systems remain constrained by their dependency on ambient lighting and potential invasiveness. As a result, recent methods has increasingly turned to alternative sensing modalities, such as inertial sensors, infrared systems, and, more recently, millimetre-wave (mmWave) radar. Radar-based HAR offers several advantages, including lighting invariance, low-power operation, and preservation of personal privacy. Nevertheless, HAR using mmWave radar faces unique challenges related to data sparsity, and noise [ 12 ] . 2.2 Data Processing in mmWave Radar-Based HAR Millimetre-wave radar, operating in the 30–300,GHz range, enables high-resolution spatial sensing through wide bandwidths and fine Doppler measurements [ 3 ] . This makes it suitable for detecting subtle human movements even in poor lighting conditions. However, radar signals often contain substantial noise and non-target reflections, necessitating data preprocessing pipelines to ensure reliable action recognition. To mitigate noise and improve tracking, recent methods have employed various data processing techniques. Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is frequently used for its ability to identify arbitrarily clusters and filter out outliers without requiring prior knowledge of the number of clusters [ 6 ] . Compared to traditional clustering methods like K-means, DBSCAN is more resilient to noisy and irregular radar point cloud data [ 13 ] . For temporal consistency and object tracking, the Hungarian Algorithm (HA) has been widely adopted for optimal frame-to-frame association of clusters or detected centroids [ 1 ] . HA provides computationally efficient solutions to assignment problems and demonstrates improved performance over alternatives such as the Auction Algorithm in large-scale scenarios [ 8 ] . Other methods have used Kalman Filtering (KF) to predict and update object states, effectively smoothing trajectories and handling partial observations in dynamic scenes [ 2 ] . In HAR applications involving near-linear human motion, the standard Kalman Filter has proven to be both effective and computationally lightweight [ 7 ] . Despite the application of these individual techniques, there has been little systematic analysis of their individual performance or how they behave in different combinations. This study therefore presents an evaluation of these methods, assessing both their recognition accuracy and computational cost, and further propose enhancements to each technique to improve performance. 3 Methodology The framework of the proposed data processing