Multi-target relative distance perception based on monocular image

These years, target detection has been frequently used in autonomous driving, UAV scene analysis and robot vision due to its technological breakthroughs. Aiming at the problem that the category and category probability information provided by target detection is insufficient in the field of autonomous driving, a multi-target relative distance perception algorithm based on monocular images is proposed. Using monocular depth estimation to obtain a depth image, apply the target bounding box information to the depth map to find the location of the target, and then combine the relative distance perception based on pixel values to get relative distance information. The results of our experiments show that the proposed viewpoint adds a new information compared with traditional target detection, can provide relative distance information, and can help to realize the assistant driving of short distance straight line.