Research And Application Of The Method To Determine The Focus Position Based On Binocular Vision

Binocular vision has always been an important area for researchers, and this is done by simultaneously acquire two images of the same scene from a different position or angle, and from this two-dimensional image can be extracted for the three-dimensional scene information. The information converted from two-dimensional to three-dimensional is a binary geometry conversion process between the three coordinate systems. Binocular vision is widely used, such as: The tracking of the scalpel in assisted surgery system to corrective action and early warning, the tracking of the surrounding vehicle in automotive safety warning systems, the tracking of moving target in a variety of scenarios.This paper analyzes the traditional concrete construction of binocular vision, and based on this the author put forward a method to achieve focus tracking use binocular vision. In this method, the first consideration is the real-time performance, and the second is the requirement for accuracy.And for these two requirements, this paper experimentally studied binocular vision system core algorithm steps, namely binocular matching technology for emphasis, thus proposed multi-scale multi-level matching method which multi-scale wavelet or cosine transform image, and cross-matched with a variety of features multi-level verification.In this paper, the traditional binocular vision system which’s access the entire image parallax, increase tracking algorithm to achieve a focus area for tracking. Compared to the conventional binocular vision system, the author do not match the whole image, but only just focus area, which greatly accelerates the matching speed while improving the match accuracy.Specifically this paper for the entire application process into several parts:(1) binocular visual image acquisition and pre-processing, in this part we describes the image acquisition and pre-processing of common methods and implementation methods, and to describe the specific implementation.(2) binocular vision geometry and calibration methods, a detailed description of the principle of binocular vision geometric model, derived the formula to get three-dimensional information; a detailed description of several calibration methods, and implementation of life in calibration, and specific calibration procedure is described and realized.(3) binocular vision focus tracking and positioning methods, this part is the central issue, the tracking and matching methods have done a detailed description; do experiments to achieve specific, and the matching method in this paper is proposed.(4) the overall implementation: from hardware to software to do a specific narrative, integrating all of the theoretical research to achieve.This paper presents a multi-scale multi-level matching method that is the image multi-scale wavelet transform and cosine transform, and cross-matched with a variety of features, next in this paper, the traditional access to the entire image parallax binocular vision system transformation, increase tracking algorithm to achieve a focus area for tracking.