DEVELOPMENT OF AN ANNOTATED DATABASE FOR ASSESING THE PERFORMANCE OF DEEP LEARNING-BASED VEHICLE DETECTION AND TRACKING MODELS

The development of a voluminous database aimed at performance evaluation of the vehicle detection and tracking algorithms is described here. The vehicle database has been created using many recorded traffic videos and annotated automatically by applying some convolutional neural network (CNN) – based object detectors. It has been split into training, validation and testing datasets and then successfully used to train, validate and test deep learning-based vehicle detectors. Some multiple vehicle detection and tracking simulations are also described. A transfer learning-based vehicle classification solution using this database and those detection and counting results is also provided here