Real-Time Vehicle Detection with Advanced Machine Learning Algorithms

Abstract

Vehicle detection not only increases road safety but also helps regulate traffic flow and improves the functioning of intelligent transportation systems. The model chosen was highly trained and tuned for an accuracy of 99%. It is far improved, considering that the detection accuracies in the literature were usually between 85% and 88%. It proposes a real-time vehicle detection system using a Convolutional Neural Network (CNN) model. The dataset “Vehicle Detection” consisted of 14,208 training images, 1,776 validation images, and 1,776 test images of size 1024x1024 pixels each and annotated with 20,000 bounding boxes. It is a 23-layer deep Convolutional neural networks model with 14.7 million parameters using ReLU and softmax as the activation functions, trained for 40 epochs using the Adamax optimiser and categorical cross-entropy loss. It used a custom callback for the hyperparameter tuning process, arriving at an initial learning rate of 0.001 and a batch size of 40. The model performed very well, with an accuracy of 99.4% on training, 99.3% on validation, and 99.4% on testing, with precision and recall of 99.3% and 99.4%, respectively.