Convolution-Based Depth Map With Shadow Removal Using Cameras for 3D Mapping in Autonomous Vehicle Driving

For autonomous vehicle driving, there are some limitations to using only 2D data so that 3D data surrounding the vehicle are used, and this supplies accurate and useful information. For 3D data, 3D mapping using a Lidar sensor is one of the methods mostly used. However, the lidar is expensive and can be easily affected by weather. Therefore, we focus on generating depth maps for 3D mapping using only cameras. Also, to remove the large amount of noise, unnecessary depth information, and uncertainty in object segmentation that occurs when using existing functions in OpenCV, we propose a depth map-generating method for 3D mapping that expresses only the information necessary for autonomous driving with less noise and clear division by convolution and shadow removal. As a result, we can create a 3D map that represents only the minimum information required for autonomous driving that increase efficiency in processing large amounts of data surrounding the ego vehicles.