Research And Development Dual-tip Bulb Detection System Based On Machine Vision

Dual-tip bulb, as a necessary part of the lighting system for vehicles, is mainly used for in-car reading lighting, license tag lighting and tail box lighting. Along with the unprecedented development of the automobile industry, the demand for dual-tip bulb is also steadily increased. In order to improve its market competitiveness, manufacturers must detect the quality of dual-tip bulbs before shipment, the detection items include whether the filament is coaxial and whether the solder joints on copper heads are qualified. At present, most of the enterprises adopt visual assessment method to do detections on such products. However, this method means not only heavy workload, time-consuming and great effort needed, but also being difficult to guarantee the consistency of product specification. Machine vision, as a rapidly developed subject, its biggest advantage is the properties of being informative, non-contact, high speed, minor error and multi-function. And the production automation level of many fields has been improved by replacing visual assessment with machine vision. This paper takes dual-tip bulb as the research object, and studies dual-tip bulb detection techniques with the application of machine vision technology, in the mean time, develops a system software suitable for bulb detection by making use of VISUAL C++ programming techniques. The research content is as following:1) Generally describe the dual-tip bulb detection system and introduce the hardware structure and model selection rules of the system based on the characteristics and the demand of the dual-tip bulb. And finally, according to the actual situation, select appropriate hardwares like cameras, lens, illuminant and acquisition cards etc, and thus design and build the experimental platform.2) Key technologies of digital images used in dual-tip bulb detections are researched, including image pre-processing technique, threshold segmentation, edge detection technique etc. Through the analysis on merits and demerits of different image algorithms,3×3 median filter is used to denoise the bulb images, and weighted average valley binary image segmentation and morphological segmentation method are used to extract the filament. This chapter focuses on discussing the edge detection techniques:in order to improve the detection accuracy, based on the traditional pixel-level Sobel operator’s coarse positioning edge, use fitting interpolation method in Gaussian curve again to make the edge location accurate, and use straight line fitting algorithm to fit edge, then linear polynomial fitting are done using least square approximant algorithm based on least square method to fit the edge. And finally design the filament coaxial decision algorithm and solder joint quality detection algorithm.3) Camera calibration is an important linkage that cannot be neglected in machine vision application. The stability and the accuracy reliability of the calibration algorithm directly affect the subsequent accurate measurement. On the basis of researches on camera calibration models, the internal and external parameters of camera are obtained by using the HAICON-Based calibration technique. This method is of high solving accuracy, and its average mean square error is 0.239546 pixels, which can meet the needs of the actual measurement.4) Develop dual-tip bulb software system. With the guidance modular design idea, the detection software system is divided into five modules including image collection module, calibration module, image processing module, pattern recognition module, display module, and software flow diagram of detection system was given. On the VS2012 development platform, a detection system that suitable for dual-tip bulb is preliminary developed by using machine vision development image base HALCON and Visual C++ Combined programming. A pretty good result of using the software that has been developed will have certain reference value for subsequent automated quality inspection.