Point Cloud-based Vehicle Monitoring Systems using Deep Learning

Point clouds produced by scanning devices such as LIDAR (Light Detection and Ranging) devices accurately maps the shape of objects and their surroundings. The use of deep learning with point cloud data focuses on object detection, semantic segmentation, and classification to accomplish safety-critical robotic perception applications such as autonomous vehicles and industrial measurement systems, requiring large amounts of computational resources and model training time. This thesis proposes efficient solutions for three-dimensional point cloud detection for vehicles, three-dimensional point cloud segmentation for indoor and outdoor spaces, and a methodology for the classification of vehicle presence utilizing compact two-dimensional point cloud sensors.