| Summary: | Estimating the sizes of large crowds in images taken from high-mounted cameras is difficult, due to large variations of crowd density at different locations, as well as perspective distortion. This research project will investigate how the latest texture analysis methods and image processing techniques in computer vision can be used to help solve this problem. In this project, we will look through multiple techniques in estimating crowd size from a distance. The first experiment estimates the crowd size on "Jacob's Method", which uses basic geometric calculation to estimate a crowd density. Using this methodology, we are able to get an estimated result but the accuracy and correctness of the result is hard to be proven. To refine the estimation of crowd density, we investigate the use of color detection and edge detection technique that can be applied to calculate and estimate the crowd size in the images. The distance of the images has to be taken from approximately 20 to 50 meters away to get a clear view of the human skin color for higher detection rate. By applying Hue Saturation Value (HSV) color-space detection technique, the system is able to detect a range of skin-tone colors. Single successful skin color detection would be counted as a blob, depending on the number of blobs; the system is able to estimate the amount of people in the image. Digital images are captured in 2-dimensional array of numbers and each number represents a pixel. Matlab has vast methods in array manipulation and a wide range of image processing libraries which is the most optimal platform for coding out the system and hence, is use as the development tool for this research project
|
|---|