3D object detection for autonomous vehicle

Abstract

3D object detection plays an important role in autonomous driving, while most state-of-the-art researches are developed based on 64-line LiDARs. However, the cost of high-resolution LiDARs are several magnitude higher than the low- resolution applied LiDARs on the makes the current research majority of low-cost robotics hard to be widely platforms. To minimize the gap between current research and real world applications as well as meet the needs of autonomous sweeper implementing which the target detection function on our is equipped with a 16-line LiDAR, in this work, traditional machine learning algorithms based on RANSAC is firstly tested. Then a image-based detector detectors are taken into experiments. methods are applied. together with six pointcloud-based After that, two data-density-based Finally, one image-based as well as two multi-modal fusion based methods are proposed in this work. All the methods above mentioned are tested on both open source dataset and self-collected NTU dataset.