Research Of Scene Depth Estimation Based On Monocular Camera And Digit Recognition Algorithm

With the development of economy and technology, the sense of safety is coming by, which sets higher request and challenge on security technique. Current security application system--Intelligent video surveillance (IVS) has shortcomings both in manual monitor and intelligent monitor. Manual monitor not only has low-level efficiency but also takes risks by manual faults. Although Intelligent monitor has achievements in several aspects, the automation should be improved in video object segmentation, object behavior analysis, object recognition and mutual process, which are burning difficulties. The depth estimation discussed in this paper is one of the most basic core technology of 3-d reestablishment, reflecting the geometry structure of object in 3-d space, which has a great of theory meaning and application value.This paper includes two parts, scene depth estimation from monocular camera based on the background of intelligent video surveillance system and meter digit recognition. The main work of this paper is following:This paper discussed several algorithms of scene depth estimation based on the monocular camera. Firstly, there exists magnification and translation in image formation process, we take image process solution and optical solution to the problem. To the depth estimation from defocus, we achieve the depth result from two images got by changing the image surface. To the depth estimation from focus, first we get images by changing the image surface using step motor, then we calculate the sharpness pixel by pixel, finally we fix the optimum focus position by advanced fibonacci searching algorithm.In addition, this article designed a meter digit recognition system based on BF518, which uses CMOS camera to collect images, stores them in BF518 and recognize them by temple and skeleton algorithm. The algorithm achieves better result to low camera resolution, images in bad environment compared with other methods.