Research And Application Of The Method To Distance Measurement Based On Binocular Vision

In recent years,with the arrival of Industry 4.0,Binocular vision in the field of artificial intelligence becomes a hot issue for researchers once again.The principle is similar with the obtaining outside information with human eyes,using two cameras to simulate human eyes,while obtaining the same scene but two positions,different angles of the digital image,the digital image acquired by two-dimensional scene information calculated to obtain three-dimensional information for this scene.As an important branch of machine vision,binocular vision in robot autonomous navigation,3D scanning,manipulator grasping,based on visual control and many other fields has a broad prospect and great application value,such as: in the surgical robot system applications can help locate surgical instrument tracking,early warning and action to correct the error,and it can help blind people to walk out of the obstacle;in cars,can help detect the distance from the vehicles in front and give the urgent judgment to keep their safe ranges.In this paper,a systematic study of the structure of a conventional on binocular vision system,from image acquisition and calibration,target detection and location,image matching,the target distance measurement research:(1)Image acquisition of binocular vision: In this paper,using two cameras with the same as models and specification.Using the VS2010 development environment based on the following Direct Show + OpenCV to achieve image acquisition and caption.(2)Calibration method of binocular vision: Detailed description of the geometric model of binocular vision in principle,the relative positional relationship between the three coordinate systems,and derived at any point within the field of spatial location.Description a comparative analysis Calibration Toolbox Matlab,OpenCV calibration API function,and the calibration process is described in detail and the code to get the final result.(3)Target detection and tracking of binocular vision: Target detection and tracking can be divided into active and passive detection and tracking of target.In this paper active target tracking use highlights of laser pointer's irradiation tracking the target(same as Mark point positioning and guidance in industrial 3D scanning).Motion detection of the target compare with traditional time-difference method,background difference,Gaussian mixture modeling method,and determine the Gaussian mixture modeling method for the detection of motion targets in this paper.Used in motion tracking algorithm CamShift track single target object and multi-target object,and using C ++programming.(4)Image matching of binocular vision: This part is most difficult and important about this paper.First introduced the image matching method in local feature,and analysis of the SIFT(Scale-invariant feature transform),SIFT has many advantages,but there is a relatively long time to matching,and it is not suitable to the real-time matching in binocular vision.We propose a new concept based on local circular moment(LCM),and give its quick calculation.And based on the LCM of binocular images,a new quickly match method is completed.It is important that we take advantage of LCM to modify the DOG(Difference of Gaussians)as a method of SIFT feature point positioning.We use the stable Gaussian gradient method or DOG operators as a local image feature to replaces the image gradient feature of SIFT method.Experimental results show that the algorithm is very stability and fast,while modify the local features of an image sampling method,with respect to the complexity of the algorithm SIFT greatly reduced,significantly improve the matching time,many experiments show that the LCM method only waste less than one-tenth of SIFT matching time.LCM can be used in real time for binocular vision target matching and positioning tracking,measurement and other techniques.(5)Distance measurement of binocular vision: The distance measurement based on binocular vision is built on the basis of binocular image matching,in this paper,using the principle of triangulation to experimental,and compared with experimental results obtained and the actual results,do error analysis.In this paper,the experimental results show that the technical ideas of this paper is reasonable and feasible,the experimental system has a certain accuracy and real-time,stability,can be very good to specific applications directly,based LCM the proposed local image feature calculation and matching,it has a certain theoretical and practical value.