Edge-based Human Action Recognition for Smart Surveillance Systems

A surveillance system is an important security aid. In most criminal cases, they become crucial evidence. However, traditional surveillance systems are incapable of preventing loss or crime. Because they only record it, unless someone keeps monitoring it. The human action recognition in this paper overcomes the limitation. Wall climbing, punching, kicking, and falling down were all recognized with a good accuracy rate in the experiment. This technology is ready to support whole-size security systems. It will send messages about detected abnormal action and the location to security staff to take action before it’s too late. Furthermore, it can be implemented on an edge-based device at a low cost and does not necessitate a large amount of bandwidth.