Weather Adaptive Model for Autonomous Driving Using Metaheuristic Algorithms

Autonomous driving has advanced in recent years, however it still faces challenges in object detection under adverse weather conditions. For self-driving cars to be safe, they must be able to detect objects correctly in difficult weather conditions. Current autonomous systems rely primarily on visual information for navigation and object detection. However, these systems may have problems when sight is impaired by rain, fog, mist, snow, or even a sandstorm. These conditions are particularly prevalent on Indian roads, where monsoons, dust storms, and dense fog in northern regions hinder detection. They might generate false positives or miss significant things, both of which could result in potential circumstances. Algorithms for visual object recognition offer an improved understanding of the surroundings. Their ability to precisely classify objects is essential for making decisions, and camera-based systems are also more affordable and used widely. This paper explores the use of metaheuristic optimization algorithms in improving the detection performance of existing object detection algorithms in adverse weather conditions. This work demonstrates the significance of hyperparameter optimization in enhancing performance of detection algorithm especially in adverse weather condition image datasets. A comparison on the relevance of updated hyperparameter on the detection performance of YOLOv9 with and without different metaheuristic algorithms like Grey Wolf Optimization, Particle Swarm Optimization is evaluated here for adverse weather conditions. The evaluation is done on datasets Detection in Adverse Weather Nature (DAWN) and Indian Driving Dataset under Adverse Weather (IDD-AW).