Vehicle Counting and Maneuver Classification With Support Vector Machines Using Low-Density Flash Lidar

This paper develops a machine-learning-based method for counting vehicles and classifying their maneuvers in a traffic intersection using inexpensive low-density Lidar sensors. First, each vehicle is automatically detected using hierarchical clustering and then its trajectory is tracked using a virtual point method that compensates for the low angular resolution of the sensor. Then, characteristic low-dimensional features of each trajectory are extracted for the classification task, so that use of the entire time-varying trajectory can be avoided. The novel features extracted include selected locations/velocities, and zero-mean singular vectors that describe the shape of the trajectory around the mean vehicle location. These features are found to provide excellent separation between various inlet-outlet maneuvers. Both simulation and extensive experimental results are presented. A single sensor that covers 2 out of 4 roads at a traffic intersection is found to work with high accuracy but has occlusion errors due to the limited coverage of the sensor. A two-sensor system that covers all 4 roads at an intersection needs an additional algorithm to merge trajectories from the two sensors to avoid double counting. High counting and classification accuracies are achieved with the developed systems.