Monocular video based object depth extraction method

Abstract

The invention relates to a monocular video based object depth extraction method, which comprises the following steps: firstly, carrying out pixel projection between adjacent key frames through using the self-calibration results of a camera so as to obtain a matching cost minimum and then obtain a locally-optimized initialized depth map; then, defining an energy function expressing color consistency constraints, geometric consistency constraints, smoothness constraints and initialization depth map information, and converting a depth map extraction problem into an energy function minimization solving problem, so that the obtained depth map is a global optimum map when the energy function is subjected to optimum solution; and then, carrying out anisotropic diffusion on the map, and then obtaining a better map segmentation result by using a Meanshift algorithm. Credible pixels in the global optimum depth map are subjected to plane fitting by using the segmentation result, thereby improving the quality of the depth map better. Meanwhile, the depth continuity of a video sequence on a time shaft is taken into consideration so as to carry out optimization on the time shaft; and finally, non-key frames are performed by using a simplifying algorithm.

Description

technical field [0001] The invention relates to a monocular video-based object depth extraction method, which belongs to the technical field of computer vision. Background technique [0002] Depth information is the main carrier of stereoscopic perception, which can play an important role in many fields such as virtual view synthesis, scene layering, multi-view video compression, and object extraction. [0003] At present, in practical applications, multi-eye cameras or depth cameras are used to directly collect depth information. This collection method has the following four types of problems: 1) The amount of data is very large. 2) The accuracy of the depth data is not high, especially the data accuracy of the depth camera drops sharply under the condition of violent movement. 3) A large number of existing precious monocular video materials cannot be reused. 4) Renewal of the industrial chain is required. The object depth extraction method based on monocular video is a ...

AI Technical Summary

Problems solved by technology

[0003] At present, in practical applications, multi-eye cameras or depth cameras are used to directly collect depth information. This collection method has the following four types of problems: 1) The amount of data is very large

2) The accuracy of the depth data is not high, especially the data accuracy of the depth camera drops sharply under the condition of violent movement

3) Existing large amounts of precious monocular video materials cannot be reused

However, multiple frames are required to be jointly optimized, which has high requirements for the continuity of frames in the scene, and at the same time, the amount of calculation is huge, which is not conducive to practical use.

Images