Jiadong Hu MSc Dissertation 2014/15

detect and track moving vehicles in traffic videos

Supervised by Y.Gotoh

Abstract

abstract

Moving objects detecion and tracking play an important role in computer vision and video processing. Background subtraction is usually used to detect object regions. But there are many problems, such as background extracing, update, illumination changes and shadows, in this method. We propose an effective moving objects detection model, which addresses the problems aforementioned. Firstly we used statistical method to build the background model then update it in real time in order to adapt to the changes of the illumination and the video itslef. After threshold operation, we utilized the morphological operation to solve the effect or disturb the noise and coonected region measurement with two parameters to process the overlapping between many objects. This article mainly aims at the following work:first, analysing and summarizing the background and significance of the topic, then summarizing the theory of point that the project involved, focusing on the theoretical formula of target detection and tracking. Finally in moving target detection and tracking , for direction , the algorithm is applied to vehicle detection project, analysis of the implementation of the system flow diagram, and gives the concrete realization of the c++ code. The project achieves the moving targets--cars tracking purposes by reading the video frames,background modeling,morphology processing, connecting component analusis, detection of target and a series of steps. Experimental results show that the purposed model achieves a better performance both in effectuveness and real-time aspect.