Joint Tracking and Classification of Moving Objects at Intersection Using a Single-Row Laser Range Scanner

Analyzing and monitoring traffic behavior at intersection is considered very important to reduce accidents, solve traffic jam and improve its accessibility. However such a data is not easy to obtain, as the visual-based systems require high sensor setting conditions, which could not be met easily in many traffic environments. In this research, we propose a novel system of monitoring and collecting a traffic data at an intersection using a single-row laser range scanner, which is set on a roadside, horizontally profiles the moving objects at the intersection at an elevation about 40cm from the ground. A method of joint tracking and classification is developed by spatially and temporally processing on laser scan data, where moving objects are classified into 0-axis object (e.g. pedestrians), 1-axis object (e.g. bicycles), 2-axis object (e.g. cars, trucks, buses). An object model is defined by picking up the typical appearances (Markov states) of an object in an instant laser measurement, and modeling each traffic flow as the transition along a chain of the Markov states (Markov chain). An accurate estimation to the shape of the object, e.g. width and length, side vectors, corner points, etc., and kinematic parameters, e.g. position, direction, speed is achieved by integrating such a partially (at each Markov State) but continuously (transitions along a Markov chain) collected knowledge to the object. An experiment is conducted in an intersection, where a SICK LMS291 is applied to collect laser data for tracking and classifying of the moving objects at the intersection, a video image is recorded to make a better understanding to the results. A ten-minute data is processed, where a successful ratio of above 95% is found by examining the results on both laser data and video images. In this experiment, although the data are processed in an off-line mode, a low computation cost (less than half of the scanning period is cost for the processing of each frame) is found of the- - method, which proved the possibility for a on-line system