Estimating Traffic Density on Roads using Convolutional Neural Network with Batch Normalization

Traffic Jam is one of the major problems of modern urban life. Regardless of the socio-economic structure of a country, almost all the countries of the world suffer from this problem. An obvious solution to this problem seems adding new and wider roads. But instead of having more roads, the cities which manage their traffic flow more efficiently are relatively good at managing traffic jams. That is why one of the best ways to tackle the problem of traffic jam can be by developing a traffic management system that can take input data from around the city and analyze that data to provide the most efficient solution to traffic. This study attempts to use computer vision along with machine learning and artificial intelligence to make a system that can determine the density of traffic from images taken by the traffic cameras mounted at the intersections. The traffic images were obtained from the API provided by Land Transport Authority, Singapore. After investigating various advanced deep learning methods, we finally built our model based on Convolutional Neural Network (CNN). Our proposed model outperformed existing models in terms of different performance metrics and achieved an accuracy of 86.56% and top 2 accuracy of 99.06%.