Development of Malaysia traffic sign detection and recognition system using neural network

Tern, Jia Chie (2024) Development of Malaysia traffic sign detection and recognition system using neural network. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full text) Development of Malaysia traffic sign detection and recognition system using neural network.pdf - Submitted Version Download (2MB)

Abstract

Road traffic signs play a crucial role by providing essential information and guidance to drivers. Nowadays, driver easily distracted especially when using navigation system, they might not be focusing and thus overlook oncoming traffc sign. Hence, this project aims to develop an Traffic Sign Detection and Recognition (TSDR) system designed for Malaysia in residential area and to analyze the efficiency of the proposed system.This project proposes the use of YOLOv3 with Spatial Pyramid Pooling (SPP) for detecting Malaysia traffic signs developed using Visual Studio Code with OpenCV library. Cross-validation is used to train the system due to small dataset used in which 800 images for training and 200 images for testing. After training, three types of validation experiments are carried out to analyze the overall accuracy of the model including the AP (Average Precision) and mAP (mean Average Precision). After testing, the overall mAP achieved by the proposed system is 93.38%, while 94.74% and 95.88% achieved in AP of Stop sign and Bump sign respectively. The system is also able to detect and identify multiple signs in multi-scaled. The results are said to be promising as the AP and mAP of the model proposed are in the range of 90.00%. In future work, system can be improved to use more variety classes of traffic sign and calculates more appropriate prior anchor boxes before training the system.

Actions (login required)

View Item