KDepthNet: Mono-Camera based Depth Estimation for Autonomous Driving
2022-01-0079
03/29/2022
- Event
- Content
- Object avoidance for autonomous driving is a vital factor in safe driving. Having the accurate depth of each barrier in the scene can contribute to obstacle prevention. In recent years, precise depth estimation systems can be attributed to notable advances in Deep Neural Networks and hardware facilities/equipment. There are several depth estimation methods for autonomous vehicles which usually utilize lasers, structured light and other reflections on the object surface to capture depth point clouds, complete surface modelling and estimate scene depth maps. However, estimating precise depth maps is still challenging. Indeed, it usually requires heavy time cost, abundant manpower and huge computing devices. Consequently, image-based depth calculation has turned into the mainstream of study and can be applied for a broad range of applications. A vast majority of camera depth estimation methods intend to determine the depth map of the whole input image using binocular cameras or a 3D camera, which is time-consuming too. In this paper, we aim to propose a novel approach that predicts the depth of the head obstacle using only a 2D mono camera. By way of explanation, the boundary boxes of barriers are extracted through a Deep Neural Network at the first stage. Rather than those methods which calculate the depth map of the entire pixels of the image, we calculated the average depth of each boundary box and set them as labels. Then labels and feature vectors (four values of the boundary box) are set as inputs of the proposed LightNet. This network maps feature vectors of the previous stage to the estimated depth values. The results suggest that the model can reasonably predict the depths of obstacles on the Kitti dataset.
- Citation
- Tavakolian, N., Fekri, P., Zadeh, M., and Dargahi, J., "KDepthNet: Mono-Camera based Depth Estimation for Autonomous Driving ," SAE Technical Paper 2022-01-0079, 2022, .