Technical and economic feasibility analysis of the implantation of a camera system with embedded processing for methods of computer vision and centralize information in cloud computing

Abstract

The objective of this project is to make possible the solution that integrates hardware with camera for image acquisition, embedded processing for application of machine learning algorithms and algorithms and neural networks with focus on the software, in order to identify automotive vehicles and license plates with letters and numbers of the Brazilian standard of vehicular identification in the images and finally to centralize the information in an infrastructure of cloud computing with auto scalability and availability. Focusing on being a service for identification and recovery of stolen vehicles, access control, monitoring and tracking for safety. This can be centralized to reduce the damage that high crime rates cause to citizens, public agencies and car insurers. Offering a high performance and scalable solution to collaborate with the development of public safety in our society. The rate of theft / robbery of cars per year in Brazil is on the order of one per minute, totaling 500 thousand per year. Meanwhile, the average value of the injury of automobile insurers is approximately R $ 40 thousand per stolen vehicle, taking into account that the fleet of insured vehicles in the country is about 150 thousand, it is possible to estimate an annual loss of R $ 6 billion a year for insurers with car theft. There are already a number of tracking solutions in the market, the most widespread is the vehicle tracker, in addition to it, insurers also already try to solve the problem on their own, allocating employees and using their intelligence to do manual search of vehicles. A third option in the market is the use of image monitoring which is what is proposed in this project. The solution consists of the installation of cameras in strategic locations where there is a large flow of stolen vehicles, with the intelligence of partner companies, the cameras are responsible for identifying and sending the license plates of the cars that circulate in the road where it is installed, if a vehicle a real-time alert is issued to the partner insurer and for the authorities to perform the vehicle recovery. The main differential of this project is the development of a 100% solution dedicated to the recovery of vehicles, while the competitors opt for comprehensive solutions for several applications, without great depth. Both the relevance of the problem and the solution were validated by the team with employees of the partner insurer. The stage of the project is the development of a product at the market level. In order to achieve this, three technical challenges must be overcome: 1) Software development and implementation with Computational Vision algorithm and methods of the team (currently Open Source codes are used) to increase the performance of our 87% , at least 95% - this number is competitive in the market. 2) Definition of the use of hardware with cameras in the market or the production of own hardware. On the one hand, market cameras can reduce logistics bottlenecks and equipment production, but they can add a higher cost to the solution because they are not dedicated components to the project requirements, and can extrapolate those needs. The technical challenge in this case should be the benchmark and test of market cameras in parallel to the modeling of resources required for the production of own hardware. 3) Implementation of a robust cloud infrastructure architecture to centralize information and has the challenge of being scalable self and available for the high volume of information in real time. The solution assumes the generation of high data (Big Data) and should be prepared to receive this contingent of information. For this, the entire solution is already developed in cloud services, such as the AWS platform. (AU)