Deep Learning based Real-time Stolen Vehicle Detection Model with Improved Precision and Reduced Look Up Time

In the recent past, the number of vehicles stolen in India has surged substantially. As per the Acko Vehicle Theft Report, in Delhi-National Capital Region (NCR), for every 12 minute a vehicle is stolen and overall Delhi-NCR accounts for more than 56% of all vehicle thefts in the country. Further, almost 3 lakh automobiles were stolen in the national capital during 2011 and 2020. While Bengaluru, India’s second-highest car theft city, accounts for 9% of vehicle theft. There exists plethora of studies for stolen vehicle tracking or detection models. The prior methods employed were time-consuming and inefficient, requiring a lot of human work to go through CCTV video recordings and trace the stolen vehicle. Some models are 96% accurate but are incapable of being employed in real time. Some are inaccurate, yet they are utilized for real-time detection. To overcome this time consuming process, a Deep Learning Based Real Time Stolen Vehicle Detection Model is proposed in this paper. Our work aims to reduce the lookup time of the vehicle in CCTV footage besides the several tasks to accomplish, like vehicle and license plate detection, optical character recognition. To achieve this, a deep learning tool known as You Only Look Once (YOLO) v8 algorithm has been used for real-time object recognition in the system. In addition, TensorFlow, an open-source machine learning platform is used for vehicle number plate recognition and for image processing. Using this strategy, we can dramatically shorten the time it takes to find the stolen vehicle. Upon rigorous experiments, results showed that the proposed deep learning based real time framework is viable and can achieve decent detection.