Research On The Algorithm Of Object Detection For Dispensers

Nowadays,high efficiency of dispensing needs the improvement of the speed in dispensing vision system.In production lines,because the actual position of each workpiece is not fixed,we must correct the position by vision-based localization technique to obtain the precise dispensing location.As machine vision becoming more and more mature,object detection develops rapidly as well and it has become one of important research subjects in dispensing field.Because of the need of high accuracy,this article proposes the algorithm of XY-axis calibration based on bilinear interpolation.This algorithm combines the methods of new XY-axis calibration and camera calibration to calibrate the dispensers.It has been proved that the errors of X and Y axes can be controlled within the scope of ±15 ?m which completely satisfies the need of dispensers.To detect the objects which have regular shapes,on the basis of original line detection algorithm,this article makes the research of rectangle detection.The new method is proposed which uses Hough transform to search the parallel and vertical line segments,and then chooses any two pair of line segments to determine whether the rectangle exists by the intersections among this four line segments.To improve the precision,this algorithm optimizes the result by the distances between any two center points of the rectangles to achieve the best rectangle.And the results of experiments proved that this algorithm has good precision,anti-noise ability,and low sensitivity as well.Besides,for complex objects,a new composite template matching algorithm based on edge feature is proposed to satisfy the requirement of high speed of object detection.A quick coarse match is conducted on the source images by interlaced search and the similarity measure of normalization.After that,a fine match is used to detect the accurate coordinate and angle in the region of interest near the center point by multi-angle templates.Compared with the traditional template matching method whose detecting time is 115 ms,the algorithm of this project detecting the same image only spends 48 ms,which is much better,so that the dispensing efficiency is substantially improved.