A Novel approach for detection and Identification of Vehicles using Single Shot MultiBox Detector (SSD) and Real Time Analytics

Abstract

With growing vehicle density and traffic, manual detection and identification of vehicles is difficult. A lot of manpower is being used for regulating the vehicle and identifying the vehicle in case of violation, blacklisted vehicles, and vehicles of interest from agencies like transport, and law enforcement. Difficult to identify the vehicles from the place and track. It requires numerous resources. Agencies spend a lot of money and effort to identify the vehicles of interest. Manually checking the databases, finding the results, and taking action require skills and is time-consuming. It also leads to delays and difficulty in taking action in real-time. With the growth of computing infrastructure and the advent of artificial intelligence, many automated solutions are proposed to identify Vehicles. In this paper, the automated solution has been proposed to create a pipeline to detect, analyze, identify, and alert the stakeholders. The proposed solution is to identify the license plates of the vehicles from locations, detect the numbers, verify the vehicle details from the database in real time, and generate alerts in case of a red flag by sending messages about the location of the vehicle and the time stamp to law enforcement officers. The research proposes an approach to use deep learning based algorithms to identify the objects from visual inputs, analyze them in real-time, and generate notifications. The solution has been implemented using the Amazon Web Services platform and infrastructure and results are captured. The experimental result shows that default SSD boxes and multi-scale feature maps yield better accuracy than Faster R-CNN with lower-resolution images.