Road Condition Detection using Adaptively Chosen Deep Learning Model and Precautionary Special-Guidance for Driver-Safety

An integrated pattern recognition model is proposed through a comparative study of image processing and various Deep Learning algorithms to surmount the fatal accidents due to potholes, broken roads, speed breakers. By processing images of the road in real-time and considering the values in the speedometer the system aims to choose the appropriate model among ResNet50, EfficientNet B7, VGG16 and AlexNet assessing the trade-off between the model-efficiency and prediction-time complexity. This adaptive selection process of DL algorithms with the vehicle's speed is adopted to emphasize the human-safety. The classified bad roads will also be flagged with a dynamic time-dependent variable associated to the number of vehicles passing through that road and the GPS locations of the road map. This integrated model not only can suggest precautionary warning in real time in the day time, but also during the night, foggy or hazy weather when visual limitation takes place.