Automatic accurate a LED new counting method
|Course||Applied Computer Technology|
|Keywords||LED die counting Mean filter Over-detection strategies Local regiondescriptors k-means clusterin|
Light-emitting diode (LED), with its advantages of long life and low power consumption, has been widely used in the areas of instructions, display and so on. However, artificial die counting is slow with low calculating accuracy and high labor intensity and sensor based die counting has the disadvantages of high equipment cost and maintenance difficulties. How to design a LED die counting method in the premise of ensuring low cost, fast response, accurate count and high robustness, is a tough problem needs to be solved immediately in this industry.The specific process of LED die counting by computer vision technology refers to completing die counting with the use of industrial cameras to get the LED chip pictures in top view and vision algorithms to process and analyze the images. The commonly used processing methods of the LED die counting are as follows:Blob analysis and sub-pixel location algorithm, the graphical contour recognition algorithm and sub-pixel linear edge detection algorithm. These three algorithms, calculating the number of the dies by the broad contours of LED dies, make the identification better and the efficiency higher in the case of a clear outline and obvious distinction of the LED dies. When taking photos of the LED chips by using actual industrial digital cameras, light impact, camera exposure or other factors will make the vertical view of the LED chips have local fuzzy, connected or deformed regions, leading to big errors when recognizing the LED dies with the algorithms mentioned above and making the precise counting difficult.This paper presents a real-time and accurate method of LED die counting. First, noise in the images is removed by mean filter; then, all the possible locations of the LED die in the LED chip pictures in top view is extracted by using a over-detection strategy; after that, local region descriptors are extracted by the location of the gray value of local area to distinguish between the LED die and image interference signal; finally, k-means clustering algorithm is used to select the target group. Experimental results show that this method can distinguish the fuzzy and connected or deformed regions, and count in real-time and accurately. This research work is mainly composed of the following contents:1. Collecting the LED chip pictures in top view and researching suitable algorithm based on image features.2. Processing the die images by using image processing technology and formatting a overall algorithm in the processes of de-noising, detection, description, and identification.3. Forming a complete system with interactive-friendly software interface which can input pictures taken by industrial cameras, display results, and support printer output.4. Testing with large-scale data to prove this method accurate.Based on the above detection algorithm, we program in the environment of Qt integrated vs2008. For gray-scale images in the size of1million pixels, the detection time is about800ms-900ms. the detection accuracy about5mm. and the average accuracy rate99%or more. Experimental results show that this detection method can count the led die in real-time, and the speed and accuracy meet the enterprise requirements. When processing and counting the LED chip image with computer, this method can greatly improve work efficiency and accuracy, and has a high practical value and promising application prospect.