Multi Traffic Scene Perception Based on Supervised Learning

Abstract: wet days, dark nights, gloomy and/or wet nights, foggy days, and many other situations with poor visibility conditions are very dangerous for traffic accidents. Current vision-based driving assistance systems are engineered to function best in temperate climates. A process called classification is used to determine the kind of optical properties that vision improvement algorithms need in order to function more effectively. A multi-class weather classification system based on numerous weather features and supervised learning is provided to enhance machine vision in inclement weather. Images of multi-traffic scenes are first processed to extract underlying visual elements, which are then expressed as an eight-dimension feature matrix. Second, classifiers are learned with five supervised learning strategies. The investigation demonstrates that the classifiers have a high recognition accuracy rate and are adaptive, and that the retrieved features may effectively capture the semantics of the image. The suggested approach offers the foundation for improving the driver's field of vision on a cloudy day and for further improving the detection of anterior vehicle detection during variations in night time illumination.

Keywords: difficult meteorological circumstances, intelligent vehicles, supervised learning, underlying visual features, and categorization