Design of vehicle detection and classification through image processing technique for surveillance system

Peong, Ying Cying (2021) Design of vehicle detection and classification through image processing technique for surveillance system. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

	Text (24 Pages) Design of vehicle detection and classification through image processing technique for surveillance system.pdf - Submitted Version Download (512kB)
	Text (Full text) Design of vehicle detection and classification through image processing technique for surveillance system.pdf - Submitted Version Restricted to Repository staff only Download (3MB)

Abstract

With the ease of technology nowadays, the surveillance system has been upgraded to do such works of gather information for analysis purpose, detect for tracking and classification of object and traffic management despite only for safety purpose. However, the conventional closed-circuit television (CCTV) is not embedded with the further processing on the video were cause to the inability to conduct the vehicle statistics analysis. Furthermore, it is non-productive to classify the vehicle by manually using the manpower. This project aims to develop an algorithm for vehicle detection and classification through the image processing technique. The developed system includes the hardware and software system which compromise of a camera, laptop and MATLAB. The image processing technique that will be used is a deep learning convolutional neural network (CNN) which is constructing by using the MATLAB. Two types of pre-trained CNN models are adopted in this project are the SqueezeNet and GoogleNet. Types of classification of the vehicle are Sedan, SUV and MPV. The developed system will be validated in terms of the accuracy, recall and precision. The result of the project is the developed system can perform the detection and classification of a vehicle with an overall accuracy of 86.7% for SqueezeNet and 97.5% for GoogleNet. The computational time per image is 0.092s for SqueezeNet and 0.194s for GoogleNet.

Actions (login required)

View Item