ระบบตรวจนับรถแบบอัตโนมัติตามทัศนภาพโดยการวิเคราะห์รูปร่าง

Abstract

This thesis proposes the techniques, methodologies and processes for vision-based automatic vehicle counting system using shape analysis. The proposed method is based on movement and shape of vehicles. The dataset used in this thesis came from the VDO Surveillance center Chiang Mai Municipality. The process is divided into 2 phases, i.e., vehicle tracking and shape analysis. In vehicle tracking, we used Subpixel Motion Estimation without Interpolation (SubME) which is divided into 2 phases, i.e. Block matching and Optical flow. The block matching and the simplified optical flow is used in the SubME. Then the Taylor series approximation is used to refine the search. The shape analysis is divided into 2 phases, i.e., Active basis and Sting grammar. Active basis model deforms an image to a template in the form of an active basis consisting of a small number of Gabor wavelet elements at each location and orientation. Sting grammar is utilized to recognize the result of Active Basis. Nearest Neighbors is then used for classification type of vehicles based on the derived String grammar