Desenvolvimento de um solução de Análise de Vídeo na Borda para Sistemas Inteligentes de vídeovigilância em tempo real

Abstract

Video Analytics has played an important role in the most varied public security sectors, mainly when applied to Intelligent Video Surveillance Systems. Edge Video Analytics seeks to migrate part of the workload of the Video Analysis process to devices close to the data source, called edge devices, to minimize problems involving transmission overhead on the network or high response latency. Therefore, this work aimed to develop an Edge Video Analytics architecture for real-time video monitoring systems, focusing on detecting and recognizing license plate characters. It proposes a division of the analysis process into functional and independent modules, implemented through Docker technology. Besides, an intermediate step between the detection and recognition steps is also proposed to discard plates with distorted identification text and prevent these plates from moving to more robust processes, as in the case of the plate recognition step or Cloud storage. For this purpose, Deep Learning Techniques were applied in a license plate classification problem according to your identification texts’ reading quality. Experiments were conducted in order to select the best algorithm for the classification model, with a better algorithm, obtaining an accuracy of 99.04% in the validation set and 100% in a small test set. In addition, a web application was developed to validate the proposed architecture, proving that the architecture meets its objectives by reducing on average 25.64% of the network traffic in its frames flow, due to a resolution quality control, and 56.65% in its license plate flow, due to the filtering step proposed.