Physical Security of Autonomous and Dumb Car

Abstract

Human activity recognition is a significant component of many innovative and human-behavior based systems. The ability to recognize various human activities enables the developing of intelligent control system. Usually the task of human activity recognition is mapped to the classification task of images representing persons actions. Autonomous cars are on their way. These intelligent and mobile agents are capable of decision making in dynamic environments.The standard car security systems respond only when you interact with the car in certain places. However, there are ways to hack the security system and all these methods require the robber to interact with the car so our main aim is to detect the interaction.Is it possible to predict unwanted humans interacting with the car? In this research we have designed deep learning based visual classifiers using monocular camera to predict unwanted interactions. These deep classifiers require lots of data. We have gathered data by attaching on-board cameras but one of the top challenge was to predict human interaction with the car from close-up shots. Although for human interaction depth cameras have high accuracy but methodology developed in this research shows that simple RGB camera can be as effective as depth cameras to recognize the interaction. This application will help in improving the security of standard cars by using cheap cameras