Unveiling Contactless Sensing with LiDAR Mobility

This paper presents a novel contactless LiDAR-based sensing method that achieves high precision even under device motion. Traditional contactless systems are often compromised by movement interference, which restricts their operational scope and continuous sensing capability. To address this, we employ LiDAR to capture depth images and convert them into point clouds for analysis. This enables the detection of subtle movements by examining changes in point cloud contours. Firstly, we adapt depth images for input into RGB person segmentation models, facilitating accurate segmentation and precise point cloud data extraction for each target. Secondly, we utilize reflections from stationary parts of the target to estimate and compensate for device movement. Our findings reveal that the transition component in the result of Iterative Closest Point (ICP) encompasses both device and subtle movements of interest. Thus, we introduce a Masked Iterative Closest Point (MICP) method that minimizes the influence of the region of interest (RoI), allowing for precise extraction of device motion information. Finally, we extract RoI motion by analyzing the differences in transition components between ICP and MICP results. Experimental validation shows that our method maintains reliable respiration sensing with a relative error below 5%, even at speeds of up to $20 \mathrm{~cm} / \mathrm{s}$ and distances up to 4 meters. This approach enhances the feasibility of LiDAR in dynamic sensing environments.