Vehicle detection and classification in difficult environmental conditions using deep learning

The time drivers spend stuck in traffic is increasing annually, on a global level. Time lost in traffic imposes costs both economically and socially. Tracking congestion throughout the road network is critical in an Intelligent Transportation System (ITS), of which vehicle detection is a core component. Great strides have been made in deep learning over the last few years particularly with the convolutional neural network (CNN), a deep learning architecture for image recognition and classification. One area in image recognition where the use of CNN has been studied is vehicle detection. This paper explores an area of vehicle detection where a little study has been made, that is the detection and classification of vehicles in difficult environmental conditions. The purpose of this paper is to build a CNN able to detect vehicles from low resolution, highly blurred images in low illumination and inclement weather and classify the vehicles in one of five classes. The final model built in this paper is able to achieve 92% classification accuracy on images in difficult environmental conditions. This model can be deployed to a smart traffic management system.