Feature Analysis and Classification Method of Road Targets Based on Two-Dimensional Point Cloud Data by Event Camera

For the future intelligent transportation, how to identify the road traffic target in real time and accurately is the basis of studying the traffic operational mode and efficiency. This paper uses a new kind of sensor Event Camera to collect data and systematically studies the road target classification method based on two-dimensional point cloud data. According to the actual traffic flow, multi-types of vehicles and pedestrians are extracted from point cloud data. Furthermore, through the distribution state of sample data, point cloud features of multi-type vehicle and pedestrian targets, including length, width, number of points, distance and aspect ratio, were extracted. Then, we analyze the characteristic data of each type of target and carry out linear divisibility analysis, which provides the basis for classification and discrimination. At last, a support vector machine (SVM) classifier is established for the target classification of small vehicles, large vehicles, non-motor vehicles and pedestrians to complete the classification of multi-type target combination. The results show that the road target classification method proposed in this paper can distinguish multi-type target combination effectively. The findings in this study can enrich and improve the existing traffic data collection system.