Research on the detection of dangerous areas such as potholes and black ice, which can cause traffic accidents on roads and bridges, helps maintain faster and smoother traffic flow and is essentially essential in protecting the lives and property of d...
Research on the detection of dangerous areas such as potholes and black ice, which can cause traffic accidents on roads and bridges, helps maintain faster and smoother traffic flow and is essentially essential in protecting the lives and property of drivers and pedestrians. In this dissertation, we propose a deep learning model that can detect dangerous regions such as potholes and black ice. To learn and evaluate the proposed deep learning model, we collect 806 images for pothole detection, 1,000 images for black ice detection. Subjectively/objectively evaluating the performance of the proposed deep learning model, we derive 95% accuracy for pothole detection and 79% accuracy for black ice detection. This study was limited to not being able to practice in real-time road driving situations, and the next study is to implement real-time output of the results with respect to the hazardous areas detected by the camera. This will also be used for research related to smart factories, smart cities and eco-friendly cars that require hard real-time such as traffic control systems, airport control systems, and satellite launch control systems in the future.