Based On The Randomized Hough Transform People Motion Analysis

Detection and follow up of a moving target is an important topic in Computer Visual. It is widely used in Security System, Intelligence Traffic, People Counting. This article shows how to process picture taken by camera using image process technology, so that some functions can be achieved, such as detection, recognition, follow up, count, etc.When camera is fixed at looking down location, it can recognizes people passing by, mainly students. Software is developed with Visual C++and OpenCV1.0, runs on Windows.After reading bulk of literature, I get a method suit for achieving this topic above mentioned. The method includes three parts. Firstly, foreground extraction. Through analysis of main trend algorithm, I select Single Gaussians Model to create and update model of background. Pre-process is done to make follow up work more efficient. It includes Shadow Erase, Morphological Filter, Edge Detection, etc. Secondly, Head Detection. For Head Detection, I choose Hough Transform which is the most traditional algorithm of shape detection. Head can seemed as a circle when looked down. So Randomized Hough Transform is proper here. The detection is better and rapider as a result of optimized algorithm. Besides, filtering is also done according to colors and profiles. Thirdly, Follow up and Count of Target. For Follow up and Count of Target, Kalman Filter is adopted after compared to Particle Filter algorithm because the movement of object can be seemed as linear movement. Model is created according to simplified Kalman Model. It contains speed, location, flags, etc. Nearest Neighbor Matching Algorithm is adopted to follow target when the direction of speed is acceptable. For counting, object is virtualized to a single target. Number increases only when target pass the Virtual Portal.Finally, I summarize the entire work and insufficiencies existed so that more research can be carried on smoothly.