With the advancement of cloud computing technologies and availability of cheaper hardware; surveillance systems are getting more reliable and accurate. Unfortunately, most of the existing systems still face a lot of limitations like latency issues, scalability issues, privacy concerns, delayed inference, reliability issues, etc. They hamper the quality and performance of surveillance systems in real-world scenarios like license plate recognition on mobile police cars, autonomous drones, etc. An extremely smart approach to tackle all such issues is to distribute some, if not all, of the computing and analytics to the edge devices. When we try to shift this computing to source of data, edge computing and artificial intelligence (for analytics) can be very helpful (especially deep learning for video-based systems). As artificial intelligence and edge computing naturally converge, their combination gives rise to the technology of the future -- Edge Intelligence. In this paper, we discuss problems with existing surveillance systems; and the use of deep learning and edge computing for the same. We introduce their combination (Edge Intelligence) as the solution to most of the major issues. We also discuss its various benefits and acceleration techniques, along with some applications in real-world scenarios along with relevant research works for further development in Edge Intelligence.