Region of Interest and Car Detection using LiDAR data for Advanced Traffic Management System

Abstract

Traffic monitoring and management have been and still continue to be a vast area of research. With the advent of different types of sensors, road transportation is becoming safer and smarter. Advanced traffic management system (ATMS) aims at improving vehicle flow and safety by leveraging technology with the traditional traffic management systems. In recent times, Light Detection and Ranging (LiDAR) sensors play an important role in ATMS. The LiDAR data, called as point cloud, consists of thousands of points in X, Y and Z coordinate. Point cloud gives a 3-dimensional representation of the surrounding. But, point cloud data is large in size and also consists of redundant information. In real-time applications, processing this substantial sized data is difficult and time-consuming. This paper proposes a complete framework for extracting the region of interest and detecting the vehicles from the point cloud, captured at traffic monitoring systems. This framework constitutes the following sub-steps: downsampling, noise removal, ground removal, object clustering and distant irrelevant objects rejection and finally, vehicle detection using the point cloud data. The MATLAB simulation of these steps suggests that this framework can be appealing to be used for ATMS. Our framework results in 64.5% accuracy of vehicle detection in the KITTI dataset.