HIF: Height Interval Filtering for Efficient Dynamic Points Removal

TL;DR

HIF introduces a real-time dynamic object removal method for 3D point clouds by modeling vertical occupancy with pillar-based height intervals and Bayesian interval filtering. A low-height preservation strategy enhances robustness in occluded regions, enabling accurate retention of static structures without ground segmentation or ray-tracing. Extensive experiments on KITTI, SemanticKITTI, and AV2 show substantial runtime improvements (6–7×) while maintaining competitive accuracy. The approach offers practical benefits for real-time SLAM and autonomous navigation, with future work aimed at adaptive pillar representations and filtering to further boost performance.

Abstract

3D point cloud mapping plays a essential role in localization and autonomous navigation. However, dynamic objects often leave residual traces during the map construction process, which undermine the performance of subsequent tasks. Therefore, dynamic object removal has become a critical challenge in point cloud based map construction within dynamic scenarios. Existing approaches, however, often incur significant computational overhead, making it difficult to meet the real-time processing requirements. To address this issue, we introduce the Height Interval Filtering (HIF) method. This approach constructs pillar-based height interval representations to probabilistically model the vertical dimension, with interval probabilities updated through Bayesian inference. It ensures real-time performance while achieving high accuracy and improving robustness in complex environments. Additionally, we propose a low-height preservation strategy that enhances the detection of unknown spaces, reducing misclassification in areas blocked by obstacles (occluded regions). Experiments on public datasets demonstrate that HIF delivers a 7.7 times improvement in time efficiency with comparable accuracy to existing SOTA methods. The code will be publicly available.

Figures (8)

• Figure 1: Comparison of FPS, accuracy and memory usage of different algorithms. The size of the points representing the algorithm's runtime memory usage.
• Figure 2: Ground point misclassification in ray-tracing. Points in (orange) represent the current scan, while points in (dark yellow) indicate ground points mistakenly classified as dynamic.
• Figure 3: Framework of the HIF algorithm. It consists of three main modules: Height Interval Construction, Interval Probability Filtering, and Low-Height Preservation. The algorithm processes an input scan $S_t$ at time $t$.
• Figure 4: Schematic of the HIF algorithm. It illustrates three point cloud scans from the same scene, labeled as A-1, B-1, and C-1. A-2 shows the height interval construction based on A-1. B-2 and C-2 demonstrate the updating process of height intervals using new point cloud scans, where green represents intervals with high static probability, and red represents intervals with low static probability. D-1 presents the combined scans from all three frames, while D-2 shows the result of filtering all points using the height intervals from C-2. Note that the parameter values in this figure have been adjusted slightly to better illustrate the algorithm's effectiveness.
• Figure 5: Comparison of different algorithms on the KITTI 05 sequence. The first two rows show the classification results for all points by each algorithm, while the last row presents the final output point clouds. Colors indicate classification outcomes: (blue) for correctly retained static points, (orange) for correctly removed dynamic points, (yellow) for misclassified static points, and (red) for missed dynamic points.