Automated smart traffic junction overhead pole camera analytics with deep neural network

Ong, Chen Yan (2021) Automated smart traffic junction overhead pole camera analytics with deep neural network. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Preview

Text (Full Text) Automated smart traffic junction overhead pole camera analytics with deep neural network.pdf - Submitted Version Download (3MB) | Preview

Abstract

As current trend in Industrial Revolution 4.0, the use of artificial intelligent technology provides a better alternative method in traffic data collection. Deep Neural Network is applied for vehicle recognition with OpenVINO toolkit in this study. Meanwhile, algorithm has been implemented for lane detection and vehicle counting based on each detected lane region. The main goal of this study is to provide reliable traffic data for traffic management system. Experiment has been carried out for 15 pre-recorded traffic videos with different time, angle view, number of lanes and weather. The performances of algorithm are then analysed and optimised based on the computation time of algorithm per frame and accuracy of algorithm. According to the results obtained, the performance of algorithm during clear sunny day is the best where it achieves 100% accuracy for lane detection algorithm and 70.62% to 87.97% accuracy for vehicle lane count algorithm. The overall performance of algorithm after optimisation has improved and all traffic videos obtained mean computation time of algorithm below than the frame rate of input traffic videos. Besides, all traffic videos had no deterioration in accuracy of optimised lane detection algorithm. At the same time, the deterioration in accuracy of optimised vehicle lane count algorithm for all traffic videos are less than 10%.

Actions (login required)

View Item