Machine vision based traffic monitoring with real time tracking

Abstract

[Abstract] : The aim of this project is to create software on a Windows platform that uses pre recorded video of trac scenes to count and give speed estimation of vehicles using real time tracking techniques. The software was developed in C using Microsoft Visual Studio 2008 and utilizes Microsoft DirectShow and the Windows GDI+ graphics library. The system utilizes background modeling and subtraction to distinguish foreground objects of interest such as moving vehicles from the static background. Object segmentation is then performed to group low level objects in the background subtracted image into blobs. Once the blobs pass certain movement criteria they are passed up to higher level tracking routines. The higher level tracking routines utilize mean shift tracking to follow the object of interest through the scene. Oriented bounding boxes are employed to provide a tighter fit around tracked vehicles and to provide visual feedback of the tracking process. The program achieved the goals of providing real time tracking, counting and speed estimation but not without numerous glitches. The software provides a bare minimum system for this type of application and needs the implementation of additional techniques such as shadow suppression, path learning and occlusion reasoning.