Fast Point Cloud Ground Segmentation Using Frequency Analysis and Linear Plane Estimation

Deploying fully autonomous vehicles requires a reliable and secure Advanced Driving Assistance System (ADAS) which, crucially, must also offer real-time performance. This study proposes fast ground segmentation for LiDAR point clouds, utilizing frequency analysis and plane estimation. The process consists of four main steps: data sampling, intensity filtering, height filtering, and ground plane estimation. Data sampling selects partial data based on the vehicle’s LiDAR installation. Intensity filtering extracts points within the most frequent intensity range, while height filtering utilizes unimodal thresholding to extract points in the dominant frequent height range, and together these filtered points estimate the plane of the segmentation boundary. Our experimental results, utilizing the SemanticKITTI dataset, have collected sampling width ratios ( $R_{w}$ ), standard deviation multipliers (K), and tolerance coefficients ( $\varDelta T$ ), which were evaluated using ANOVA. The optimal parameter combination yielded 91% accuracy, 84% precision, 94% recall, and an F1 score of 89%, all taking 2.7 ms of processing time.