Real-time traffic light violations using distributed streaming

Vehicles controlled by intelligent technologies, whose goal is to reduce human error and ease congestion, do not solely rely on human resources. Cities worldwide use camera systems to monitor the traffic, which collects the images and processes them through different computer vision algorithms. It is challenging for traffic monitoring systems to maintain their accuracy during the day and night lighting conditions, camera location relative to objects, video quality, traffic light position relative to the crossing line, and object angle from the surveillance camera. In this paper, we propose an improved traffic light violation detection method that concurrently streams videos through Apache Kafka and processes them with Apache Spark. It continues to operate for long periods without human intervention and adjusts automatically to changes in the environment. The violation detection algorithm utilizes a modified YOLOv5 and Hough space to efficiently capture the violation. YOLOv5 is a lightweight, fast, and efficient algorithm for real-time object detection. The improved YOLOv5 retrieves the object coordinates relative to the traffic lights, and Hough space analysis is employed to determine the violation region during the red traffic light. Hough space considers the object’s location and angle relative to the traffic lights. The model performs well in various situations of the input video datasets, as validated by performance metrics. The outcomes of extensive experiments show that the approach is well suited for deployment in real-time traffic violation detections. The outcomes are compared to a number of performance measures for object identification and traffic violations. In terms of traffic light violations, the model had 88.24% accuracy. The model is scalable enough, and it can deal effectively with real-world traffic video data at large scales.