Proposed Model for Detection and Classification of Vehicles in Real-Time Video Based on Deep Learning
Dhuha J. Jawad, Raheem Ogla, Abdul Monem S. Rahma · 2022
Many people die every year due to fatal accidents that occur all over the world as a result of the increase in the number of vehicles on the roads, the most common reasons being driver inattention or poor vision in bad weather conditions. Therefore, human detection and classification of vehicles is complex. Due to the development of the automobile industry due to technological advancement, it has become critical to employ self-driving cars since the computer system in them assumes all driving tasks in order to avoid collisions and accidents. However, there are numerous issues with an autonomous car's capacity to recognize and track vehicles. One of these issues is the wide variety of vehicle sizes and shapes. Our paper will focus on resolving this issue, as this paper proposed a hybrid deep learning model for vehicle classification and determined the location of vehicles based on the principles of the YOLO technique. This study uses two datasets in training and testing: The Stanford car dataset and three classes from the open image dataset v6 (bus, truck, and motorbike). The proposed model achieved high results in vehicle detection and classification, with an accuracy rate of 98.19%. The proposed model has been a success as it was run for the purpose of detecting and classifying vehicles in many videos in real time and under different environmental conditions.