Cargo Vehicle Classification System Through Axle Detection

A system capable of detecting trucks and their axles using convolutional neural networks (CNN) is presented. First, the system detects the trucks in an image by defining a bounding box around it and then proceeds to count the axles of the truck. The selected CNN models used for evaluation are the Faster R-CNN and Inception-ResNet V2 due to its performance according to the reviewed literature. For training and testing of the proposed solution a mobile unit was implemented to collect an image dataset consisting of trucks and other vehicles in a high-volume highway. Performance for this application was evaluated using different metrics resulting in a score of at least 92% in F1 Score, Recall and Precision.