Outdoor monocular image depth estimation method based on structured random forest

Abstract

The invention discloses an outdoor monocular image depth estimation method based on a structured random forest, and belongs to the field of depth estimation. The method includes: using features of image blocks to perform classification; giving the same depth to the image blocks with similar features, predicting the image depth by adopting a structured random forest, discretizing the depth structure label of the scene by each node through information gain, and then carrying out classifying until the similarity of the image blocks in each leaf node in the tree reaches a certain threshold value;finally, combining the results of each local block to form a complete depth map prediction result. Reliable depth clues can be obtained through multi-aspect features, the structure and local information of a scene can be well considered through partitioning, and the accurate and reliable absolute depth is estimated through existing depth information. Through multiple times of discretization of thedepth structure labels at the nodes, the structures of the depth blocks are classified, depth estimation is facilitated, and the precision of the estimated depth is better improved by randomly combining the results of multiple trees.

Description

technical field [0001] The invention belongs to the field of depth estimation, and more specifically relates to a method for estimating the depth of an outdoor monocular image based on a structured random forest. Background technique [0002] Compared with indoor images, the use of depth equipment to capture outdoor scenes is more affected by the environment. Outdoor atmosphere, light, heavy fog and other environments will have a greater impact on depth acquisition equipment. Depth devices based on structured light such as Kinect cannot be used outdoors, and the cost of devices such as lidar that can be used outdoors is too high, and the use of binocular camera stereo matching methods to estimate depth has a large amount of calculations. The low area effect is not good. In addition, self-driving cars, SLAM and other fields have great requirements for the three-dimensional structure of outdoor scenes, so monocular depth estimation based on outdoor images is a problem worthy ...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the defects of the prior art, the purpose of the present invention is to solve the technical problem that the depth accuracy predicted by the prior art is not high

Images