Reconhecimento automático de armas de fogo no interior de veículos.

Abstract

The increase in urban violence in Brazil has highlighted the growing number of armed robberies inside vehicles. The data recorded impressed, adding to the alarming number of 57 vehicles robbed per hour in the country. Manaus presents itself as one of the Brazilian cities with the highest number of vehicle assaults. The Sinetram Passenger Transport Companies Union (Sinetram) already registers in the first four months of 2017 the alarming number of 1,120 bus robberies in Manaus. On the other hand, the Union of Taxistas do Amazonas (Sintax-AM) points out that at least ten taxi drivers are assaulted per day in the city. The objective of this work is the development of a method that automatically recognizes armed robberies inside automotive vehicles. Contributing in this way to the fight against urban violence and enabling the quicker and more effective action of public security agents. The approach that will be adopted is the creation of a set of local descriptors, generated from a sequence of images of firearms (revolvers and pistols). These descriptors provide an information base capable of identifying the presence of firearms in the images captured from the interior of walking vehicles. This approach dispenses with the location of the weapon in the image space and recognizes it from a set of optimized features. The obtained results show that the developed method recognizes the firearm in different situations of movement, with hit rates above 80 % in all metrics used. The method is integrable with modern vehicle safety systems and robust enough for continuous monitoring of the interior of pscsenger cars.