Monocular depth estimation method based on target surface point height and uncertainty

The invention discloses a monocular depth estimation method based on the height and uncertainty of a target surface point, and proposes to replace the height of an object with the physical height and projection height of the surface point of the object so as to obtain a plurality of depth candidate values. In addition, the uncertainty of the height is estimated, and the depth of the high-uncertainty inference instance is replaced with the depth of direct regression. And combining the depth predictions according to the uncertainty to obtain the final target depth. According to the method, the best performance is obtained on the KITTI data set, and the reasoning time meets the real-time requirement of automatic driving. 本发明公开了一种基于目标表面点高度和不确定性的单目深度估计方法，提出用物体表面点的物理高度和投影高度来代替物体本身的高度，从而得到多个深度候选值。此外，估计高度的不确定性，并用直接回归的深度代替高不确定性的推断实例深度。根据不确定度，将这些深度预测结合起来，得到最终目标深度。本发明在KITTI数据集上取得了最好的性能表现，并且推理时间满足自动驾驶实时性的要求。