| With the "Made in China 2025" concept known by the general public, the machine vision industry has developed rapidly. Machine vision has many features such as non-contact, high efficiency, easy to operate, good stability and high precision.It was widely used in flexible automation field such as license plate recognition, two-dimensional code recognition, detection of industrial product classification, tracking moving objects flexible automation field. This paper's resource was from the cooperative development projects with a large domestic production enterprise that the author involved in, mainly in order to solve the corporation's long existed problens of low degree of automation,inefficent artificial recognition and classification.Therefore,a set of hardware platform and software system for automobile wheel hub identification and classification that based on multi feature fusion was established,including hardware design, software algorithm design and human-computer interaction interface design.The main research steps are as follows:According to the design requirements of the automatic identification and classification system for the wheel hub that could be applied to the scene, firstly the research object and the system function were analysised to summarize the technical index.secondly, the technical index and system function were analyzed, and the whole system solution, hardware solution and software solution were established seperately. Thirdly Complete the establishment of the black box system and the selection of hardware including light source, camera obscura, IPC, PLC, industrial cameras, lenses etc. In order to reduce the influence of light change on the image stability, the brightness compensation system based on the intensity of light was added to the black box.The software algorithm has three steps:image pre-processing, feature extraction, classification to determine.The first step was image pre-processing including grey scale processing, histogram equalization, bilateral filtering and binaryzation etc of the obtained image.Besides,for the purpose of seperating hub from its background, this paper tested on the common image reduction algorithm and found that it is insufficient to meet the actual testing requirements.In order to solve this problem, a new background removal method based on hub was proposed and the experimental results was fairly good.In order to make the profile of the wheel hub more brighter, we carry out morphological operations such as corrosion and expansion. The canny edge detection operator was used to extract the edge information of the wheel hub, and the Freeman coding method was used to search the pixel point of the edge contour, then the least square method was used to fit the obtained pixel coordinates. The edge information of wheel hub that were extracted include the hub diameter, web axis number, spokes, wheel hub to upper edge distance, a hub to lower edge distance, void area ratio, Hu moment feature.Next step was to design the classifier, the feature data were normalized and recogized by Euclidean distance method, according to the principle of software design, finally the human-computer interaction interface of this project was designed, and the use steps was introduced. The on-site verification of the system and a month-long commissioning at the scene were implemented. The tests of this system has been successfully identified more than 80 types of wheels,with higher than 96% recognition accuracy rate and close to zero misclassification rate.lt indicate that the classification results was sufficient to meet the requirements of visual inspection in industry field. |
PDF Full Text Request