Embedded Raspberry Pi Vehicle Detection and Classification using Single Shot Detector and Dilated Multi-Column Convolutional Neural Network

Detection and classification of vehicle types is one of the most important applications in field of road safety. In this paper, we propose a deep learning process to detect and classify vehicles by using Single shot Detector (SSD) for detection and Di-lated Multi-Column Convolutional Neural Network (DMCCNN) for classification. Rather than using a fixed-scale convolutional layer, the second model we use is capable to extract features from various scales of an image employing different dilated filters to improve the performance classification, especially for similar-looking vehicles like Suv and Sedan. An embedded Raspberry Pi vehicle detection and classification system is developed using a built-in Pi camera. The results are comparable with desktop-based results in the literature yielding an accuracy of 95.93 % on BIT dataset.