Research Of Dynamic Detection Of Workpieces Based On Image Recognition Technique

With the development of photoelectricity and digital image process technology, the field of the industrial detection has cast aside traditional manual detection, beginning to adopt online detection of workpieces using automatic recognition technology. Compared with traditional way, automatic recognition system can not only recognize automatically but also can transmit data to computer or controller in real time. Data will be processed in time which will enhance the level of automatization of industrial production. In this paper, an on-line detection technique of workpieces is introduced in detail. A fast detection algorithm based on the gray feature is discussed and improved, therefore, the detection of the workpieces states can be completed in real time. In contrast with other detection system, it is accurate to say that the system discussed in the paper is efficient, fast and reliable. There are two processes of the online detection system of workpieces: process of learning and process of recognition. In the process of learning, we firstly get an image by taking pictures of the accessory using a normal vidicon, extract the characters secondly. In the process of recognition, a picture of moving workpieces is taken and the features are extracted, then we can carry out classification using predefined discrimination function and reporting the state of the system. The experiments are conducted under the nature light and the images are very coarse, therefore, it is necessary to have the images processed in advance. In this paper, the image graylization process and the 3×3 median filter are adopted to reduce coarse and stand out the structure of the accessories. The images which processed in advance are ready for the classification. The traditional technology of image process adopts the verge pick-up means, then pick up the characters of the images. In this paper, the system requires the detection going along in real time. In view of the spending of the time and the unsatisfied effect, the verge pick-up is not adopted in this paper, however, a speedily detection arithmetic based on the gray characters is introduced in the paper, using the gray characters to detect the objects directly. The algorithm recognizes the objects based on the gray feature in the pixels. For a simple object, the gray distribution has coherence to some extent, it is to say that the most of the gray of the pixels are in a given range and the gray of the background pixel out of the range as a rule. This given range can be described by one or a few thresholds. Based on these features, the algorithm can detect simple objects, but the detecting effect of the complex objects is not so ideal. By researching the BMP files of the objects, it is found that there are some heavy undulations on the verge of the object and the inside. The undulations can be described by gray apex and valley. When it is detecting a complex object, the gray apex and valley can be a gist to detect the objects. The effect and the fastness are both ideal. The steps of the characters pick-up are as follows: 1. Figure out the gray threshold of the object and background pixel S 1. 2. Figure out the threshold of the gray apex and valley S 2. 3. Pick up the features. Including the simple objects and the complex objects. For the simple objects, the fast detection arithmetic based on the gray features is adopt, adopting threshold S 1. Define C1 ( k )(0 < k≤176) as the numbers of the pixels which gray is less than S1 . Define eigenvalue Min1 as the minimum of C1 ( k ) in the object region and Max1 as the maximum. Define eigenvalue T 1 as the C1 (l ) (0 < l≤176)in the first line. For the complex objects, the improved detection algorithm based on the gray features is adopted, adopting threshold S 2. Define C 2( k ) (0 < k≤176) as the maximum of the neighboring pixel gray difference which gray is less than S 2. Define eigenvalue Min 2 as the minimum of C 2( k ) in the object region and Max1 as the maximum. Define eigenvalue T 2 as the C 2( k ) (0 < l≤176)in the first line. The steps of the classification are as follows: a) When C1 ( k ) is greater then Min1 and less then Max1 , meanwhile, the error with T 1 is less then five pixel, it is granted that the right object is detect. b) When C1 ( k ) is greater then Min1 and less then Max1 , meanwhile, the error with T 1 is greater then five pixel, it is granted that the wrong object is detected .