Towards Autonomous Driving: A Machine Learning-based Pedestrian Detection System using 16-Layer LiDAR
Stefan Mihai, Purav Shah, Glenford Mapp, Huan Xuan Nguyen, Ramona Trestian · 2020
The advent of driverless and automated vehicle technologies opens up a new era of safe and comfortable transportation. However, one of the most important features that an autonomous vehicle requires, is a reliable pedestrian detection mechanism. Many solutions have been proposed in the literature to achieve this technology, ranging from image processing algorithms applied on a camera feed, to filtering LiDAR scans for points that are reflected off pedestrians. To this extent, this paper proposes a machine learning-based pedestrian detection mechanism using a 16-layer Velodyne Puck LITE LiDAR. The proposed mechanism compensates for the low resolution of the LiDAR through the use of linear interpolation between layers, effectively introducing 15 pseudo-layers to help obtain timely detection at practical distances. The pedestrian candidates are then classified using a Support Vector Machine (SVM), and the algorithm is verified by accuracy testing using real LiDAR frames acquired under different road scenarios.