Unsupervised monocular view depth estimation method based on multi-scale unification

Abstract

The invention belongs to the technical field of image processing, and discloses an unsupervised monocular view depth estimation method based on multi-scale unification, and the method comprises the following steps: S1, carrying out the pyramid multi-scale processing of an input stereo image pair; s2, constructing a network framework for encoding and decoding; s3, transmitting the features extracted in the encoding stage to a reverse convolutional neural network to realize feature extraction of input images of different scales; s4, performing unified up-sampling on the disparity maps of different scales to an original input size; s5, performing image reconstruction by using the input original image and the corresponding disparity map; s6, constraining the accuracy of image reconstruction; s7, training a network model by adopting a gradient descent method; and S8, fitting a corresponding disparity map according to the input image and a pre-training model. According to the method, networktraining does not need to be supervised by using real depth data, the binocular image which is easy to obtain is used as a training sample, the obtaining difficulty of network training is greatly reduced, and the problem of depth map holes caused by low-scale disparity map blurring is solved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an unsupervised monocular view depth estimation method based on multi-scale unification. Background technique [0002] With the development of science and technology and the explosive growth of information, people's attention to image scenes is gradually shifting from two-dimensional to three-dimensional. The three-dimensional information of objects has played a great role in daily life. Among them, three-dimensional information is the most widely used The most important thing is the assisted driving system for driving scenes. Due to the rich information contained in the image, the vision sensor covers almost all relevant information required for driving, including but not limited to lane geometry, traffic signs, lights, object position and speed, etc. Among all forms of visual information, depth information plays a very important role in driver assistance systems. For...

AI Technical Summary

Problems solved by technology

People use the powerful learning ability of the neural network to design various models to fully mine the connection between the original image and the depth image, so as to train the scene depth that can be predicted according to the input image, but as mentioned above, the real depth information of the scene It is very rare, which means that we need to break away from the real depth label of the scene and use an unsupervised method to complete the depth estimation task

One of the unsupervised methods is to use the timing information of monocular video as the supervisory signal, but this kind of unsupervised depth estimation method uses the video information collected during the motion, so the camera itself has motion, and the image sequence The relative pose of the inter-camera is unknown, which leads to the need to train another pose estimation network in addition to the depth estimation network, which undoubtedly increases the difficulty of the already complex depth estimation task.

In addition, due to the scale uncertainty of monocular video, this method can only obtain relative depth results, that is, only the relative distance between pixels in the image can be obtained, but the distance from the object to the camera in the image cannot be obtained

In addition, the method of unsupervised depth estimation has the situation that the texture of the depth map is missing or even empty due to the blurred details of the low-scale feature map, which directly affects the accuracy of depth estimation.

Images