| Summary: | 碩士 === 國立高雄應用科技大學 === 資訊工程系 === 97 === At present, there are many people counting on video processing literature, the crowd of non-crowded circumstances, may have a good accuracy of the count, but in crowded circumstances, the accuracy will drop. In this paper, also in the crowd have a higher accuracy of the method.
This paper presents an automatic method of counting the crowd, set up at the gateway through the top, and vertically down the lens of the camera according to the perturbation of the screen to retrieve information, the use of image processing and machine vision approach to image capture pixels from the crowd using the algorithm proposed in this paper to analyze, thus the function of counting.
The main technique is the first to use the frame difference method of the film to detect the edge of the crowd moved out, and then, in the detection of images to do dilation of morphological processing , after the use of morphological methods to detect the outline, and then use region growing to be the outline of the circuit, to fill the region, after the image to fill the hole is still there. In order to improve the hole for the second procedure to fill to cut out moving objects. Then, the characteristics of individual image capture, on the part of interest to be analyzed, would be inconsistent with personal images of people eigenvalue images, re-cutting and re-analyzed, followed by images of people tracking and counting, in order to achieve automation of the counting function of the crowd.
Finally, we will experiment with the manual count to compare the results, the system in normal circumstances, its precision can reach about 85 percent.
|
|---|
