Vehicle Detection and Classification in Traffic Images Using ConvNets With Constrained Resources

This paper proposes a convolutional neural network model for detection and classification of vehicles present in digital images into six categories, namely Bus, Microbus, Minivan, Sedan, SUV, and Truck. Experimental results with the BIT-Vehicle Dataset report a mean average precision (mAP) of 92.40% and the average intersection over a union (IoU) of 81.26% with the average inference latency under 70 milliseconds in an environment equipped with a graphics processing unit. We conclude that the model is discriminative and capable of generalizing the patterns of the vehicle type classification task while not requiring expensive computational resources. These features suggest that the model can be useful in the development of embedded intelligent traffic systems improving accuracy and decision latency.