Search

Moving-object tracking (estimating position and velocity of moving objects) is a key technology for autonomous driving systems and driving assistance systems in mobile robotics and vehicle automation domains. To predict and avoid collisions, the tracking system has to identify objects as accurately as possible. This paper presents a method for recognizing vehicles (cars and bicyclists) and people using a 64-layer ground laser scanner. When laser-scanned data are captured by the laser scanner, laser-measurement points related to objects are extracted by the background subtraction method and are clustered. Eight-dimensional features are extracted from each of clustered laser-scanned data, such as distance from the laser scanner, velocity, object size, number of laser-measurement points, and distribution of the reflection intensities. The machine learning methods (support vector machine (SVM) and random forests (RF)) are applied to classify cars, bicyclists, and people from these features. The experimental results using “The Stanford Track Collection” data set show that the classification accuracy using the SVM-based method is higher than the RF-based method. In addition, they show that the use of the proposed eight-dimensional features provides better classification accuracy and shorter processing time than the use of the conventional 26-dimensional features.