An Adaptive Traffic Light Control System Based On Artificial Intelligence

Abstract:The phenomenon of traffic congestion is one of the most widespread phenomena in the world, especially in developed countries where the number of vehicles is increasing, which means that many traffic jams occur in the streets, especially at times when employees go to and from work, which is known as peak times and is concentrated usually in the early morning, and this gives a great feeling of annoyance and boredom as a result of wasting hours of time in the streets, and being late for work because of not being able to reach it at the appropriate time. Add to that a signal system traditional traffic, which played a prominent role in these bottlenecks, can no longer smooth management and organization of traffic, because it did not develop with the development of cities. From this point of view, studied and applied, a system based on image processing that uses the technology to detect objects in the image by removing impurities, and thus determining the traffic density based on the number of objects. In the picture, which represents the vehicles, the traffic lights were controlled depending on the traffic density on each road. The proposed system is implemented using MATLAB R2021 language, and for performance evaluation, classification accuracy is used as the evaluation metric. On traffic lots (MTID) the data set is frames of video recorded at 40000 (frames per second) FPS and taken by a drone. The Multi-view Traffic Intersection Dataset is used in the proposal system, it is available on "https://www.kaggle.com/datasets" . Categories: Bicycles, Cars, Buses, Trucks. To test the effectiveness of the proposed approach. Three pre-trained networks (AlexNet, ResNet, and DenseNet) were implemented. Using the ResNet50 model resulted in a significant improvement in performance. In particular, the proposed system using ResNet50 achieves an accuracy average of 99.47% on the (MTID) datasets respectively.