Sensor-Driven Autonomous Vehicles: Junction Safety using AI and Obstacle Avoidance

The demand for autonomous vehicles (AVs) has surged in recent years with the advancements in Artificial Intelligence (AI). AI, sensors, and real-time processing have made autonomous driving more viable, especially in ensuring road safety at complex traffic junctions. The traffic junctions pose significant challenges in metropolitan areas for all kinds of autonomous and non-autonomous vehicles. At such overcrowded junctions, precise detection of traffic signals, accurate measurement of surroundings, and timely decisions are necessary to avoid inconvenience. This paper presents a novel system designed to enhance junction safety by employing a combination of sensors and AI algorithms. The proposed system involves: i) LIDAR, for precise distance measurements ii) Camera, for visual recognition of traffic signals iii) YOLOv5 (a CNN-based object detection model), for accurate interpretation of sensor data. Our approach leverages transfer learning, utilizing pre-trained models to boost the accuracy of traffic signal detection while reducing training time. The proposed method enhances the system’s ability to recognize traffic signals and detect obstacles in real-time, even in diverse lighting and weather conditions. The integration of LIDAR with neural network-based decision-making not only improves obstacle detection but also ensures reliable traffic light recognition. The proposed solution is rigorously tested in both simulated environments and real-world conditions, showcasing significant improvements in safety measures compared to conventional AV systems.