Vehicles Detection and Tracking At Night based on the Light Diffusion Model and the Optical Flow Algorithm

• Main Authors: Jun Ting Wu, 吳俊廷
• Other Authors: J. D. Lee
• Format: Others
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/sh248b

碩士 === 長庚大學 === 電機工程學系 === 102 === Vehicle detection is a common function in today’s ADAS ─ advanced driver assistance systems. Vehicle detection plays an important role in alerting the drivers on whether the environment in which they are driving is safe or not. It collects information for each car near the host vehicle, and alerts the driver whether each vehicle belongs to a highly risk category, or a low risk, safe and normal car after a careful analysis the collected data. Usually we include information like colors, distance, textures or shadows for analysis, but these information are only available during the day time. Once the environment is changed into the night time, the availability of these types of information disappears because the lack of luminous beam. So we instead use lighted vehicle lamps as information to detect vehicles in the night time. In this thesis, we use the light source diffusion model to establish a diffuse light intensity map, in which high intensity regions are taken as the locations of the vehicle lamps. And we also use the optical flow algorithm to tracks the lamps in real-time. In summary, the following tasks have been accomplished in this thesis. (1) Build a light intensity map based on the light source diffusion model. (2) Analysis the light intensity map to find the distribution of the vehicle lamps. (3) Using the center mass of ROI’s as the input to the optical flow algorithm to distinguish the differences between them. (4) Putting the collected data into a fuzzy system which analyses the possibility of the existence of vehicle by applying the fuzzy rules to the relationships between the lamps. The proposed method can detect vehicles by using information from vehicle lamps. Also this method can decrease the interfering effects of non-vehicle light sources. This method has been tested under real-time situations.