Vehicle Shape Matching & Recognition for ADAS Using Tchebichef Moment

Tan, Zhi Zin (2017) Vehicle Shape Matching & Recognition for ADAS Using Tchebichef Moment. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Tan Zhi Zin Full text.pdf Restricted to Registered users only Download (1MB)

Abstract

Vehicle detection from images gotten from a moving vehicle is an important application for advanced driver assistance systems as well as autonomous vehicles. The focus of this work is about the edge detection, feature extraction and classification for the detection of rear-view vehicles. The vehicle detection algorithm can be divided into two categories which are hypothesis generation and hypothesis verification. Under hypothesis generation category, possible locations of vehicles are extracted from the image while a high false detection rate is maintained. The possible location of vehicles is detected by separating Sobel edge detection into both vertical and horizontal components. For hypothesis verification, it is done in two steps which are feature extraction and classification. Tchebichef Moment (TM) is used as the appearance based feature extraction method to extract the features of a vehicle and act as input for the Neural Network (NN) as well as Support Vector Machine (SVM). NN and SVM are two types of power classifiers and are trained to classify the vehicles and eliminate the falsely detected vehicles generated in the first category. Experimental results show the comparison and that the algorithm proposed is able to maintain a high detection rate of vehicles under normal condition.