Classification of Objects Based on the Construction of an Enveloping Polygonal Chain for the Highly Automated Vehicles

This paper presents a method for obstacle recognition and classification using LiDAR data to enhance the operation of highly automated vehicles (HAVs). The approach focuses on constructing an enveloping polygonal chain around a point cloud representation of an obstacle detected by a 3D-LiDAR sensor. The polygonal chain is built using the highest points of the object, and each segment's height difference and angle of inclination relative to the horizon are analyzed. This analysis enables the classification of detected objects into surmountable and insurmountable obstacles, facilitating more efficient and safer route planning for HAVs. Unlike computationally intensive machine learning techniques, this method emphasizes geometric analysis for object classification, ensuring reduced computational overhead and faster processing. Experimental validation on a GAZelle vehicle equipped with an LSLIDAR CH-64 sensor demonstrates the accuracy and reliability of the proposed approach across various obstacle types.