Non-Calibration Sensor Fusion Using High Resolution 3D LiDAR

Light detection and ranging (LiDAR) and camera are the most commonly used combination in sensor fusion techniques. In this paper, we propose a method to recognize obstacles by fusion of images and point cloud data generated from high-resolution 3D LiDAR. By applying the image to a deep learning network, we obtain inference information about the obstacle, and based on this, we convert the pixel location of the obstacle in the image into a mapped point cloud. This enables more sophisticated obstacle recognition by converting 2D data into 3D space data. The algorithm is implemented in an autonomous vehicle and tested at the proving ground of the Korea Intelligent Automotive Parts Promotion Institute (KIAPI), under the 2024 autonomous driving competition for university students, Daegu, Republic of Korea