A Supervised Data-Driven Depth Estimation Method for Monocular Video

Abstract

The invention discloses a monocular video depth estimation method driven by supervised data, comprising the following steps: (1) acquiring sample video sequences and corresponding depth sequences as training data sets; (2) using tracking-based superpixels The segmentation method divides the training data set and extracts the features of each segmentation unit; (3) constructs a network model combining convolutional neural network and spatio-temporal conditional random field; (4) uses the training data set, segmentation results and corresponding The spatio-temporal convolutional neural network field model is trained; (5) the video sequence to be estimated is segmented, and the features of each segmentation unit are extracted; (6) the video sequence to be estimated, the segmentation result and the corresponding features are input into the trained model , get the depth sequence. The invention takes into account the consistency of time and space and the accuracy of hierarchical relations, and improves the quality of monocular stereoscopic video.

Description

technical field [0001] The invention belongs to the technical field of pattern recognition, and in particular relates to a supervised data-driven monocular video depth estimation method, which is used for automatically estimating the depth value of a video sequence from monocular video. Background technique [0002] With the development of science and technology, 3D movies and virtual reality are enriching people's lives. However, whether it is the 3D movies that have been popular all over the world or the virtual reality that is currently in the ascendant, there is a serious problem, that is, the current lack of 3D resources. Therefore, predicting depth through monocular video and then obtaining binocular stereoscopic video through viewpoint synthesis has become the main method to solve the current shortage of 3D resources. [0003] In this technical approach, the depth estimation of monocular video has been widely concerned by researchers as an important part of it. The ...

AI Technical Summary

Problems solved by technology

However, the current depth estimation method based on machine learning mainly predicts a single image. If it is directly used in the monocular video depth estimation task, the temporal and spatial consistency of the prediction results is poor, which cannot meet the needs of users.

Images