Method and device for detecting lens in real time in fully automatic two-dimensional (2D) to three-dimensional (3D) technology

Abstract

The invention discloses a method for detecting a lens in real time in fully automatic two-dimensional (2D) to three-dimensional (3D) technology. The method comprises the following steps of: reading a current frame; partitioning the current frame into a plurality of sub-blocks; counting inter-frame difference values and accumulated frame difference values of the plurality of sub-blocks relative to a previous frame; dynamically adjusting corresponding low threshold values and high threshold values of the plurality of sub-blocks according to the inter-frame difference values of the plurality of sub-blocks; and obtaining a lens detection result of the current frame according to the inter-frame difference values and the accumulated frame difference values of the plurality of sub-blocks and the corresponding low threshold values and high threshold values. The method has the advantages of giving the lens detection result of the current frame on line in real time, achieving a high recall ratio on the basis of ensuring a precision ratio and basically meeting the requirement of subsequent processing.

Description

technical field [0001] The invention relates to the field of computer multimedia technology, in particular to a real-time lens detection method and device in fully automatic 2D to 3D technology. Background technique [0002] A shot in a video refers to a sequence of frames that are continuously shot by the same camera and represent a continuous action in time and space. From the perspective of context semantic content, shot is the basic unit of video sequence content. For a real-time fully automatic 2D to 3D system, a common process is to divide it into three steps: lens detection, depth calculation and DIBR rendering. Among them, shot detection requires the ability to simultaneously detect the boundaries of shot changes and shot gradients in the video; depth calculation is based on shot detection, and for adjacent image frames in the same shot, it is processed by block matching algorithm and bilateral filtering. , to obtain an approximate depth map; while DIBR rendering r...

AI Technical Summary

Problems solved by technology

For example, some methods have too high computational time complexity, they may need to convert the original image from RGB space to HSV or La * b * color space, or need to perform block matching for each sub-block of the image, or need to detect and compare the edge of the object in the image, so all in all, they will consume a lot of time when performing these processes, and cannot meet the full Real-time requirements of automatic 2D to 3D system

In addition, the principles used in some methods are too complicated and the steps are too cumbersome, which is not conducive to the realization of the hardware of this module in the future.

However, some clustering-based shot segmentation methods need to read the information of all frames first, and then classify these frames, which is not in line with the online processing of a fully automatic 2D to 3D system that reads a frame and judges a frame. model

Images