Since the CCTV cameras been introduced in this world, society has started to depend heavily on the usage of this technology for the high security purposes in most of the public and private areas. It is convenient to use these CCTV footages in courts as evidence and has been beneficial many times. But these footages are given priority and checked later when the incident has already taken place and that too after some period of time and not in real-time of happening. The screening of the multiple CCTV footages on a single monitor is done with very less efficiency as the ratio of number of CCTV footages to that of number of surveillance staff is very high. Also, the human unreliable supervision due to many reasons like tiredness from physical or mental effort, worker boredom, or discontinuous observation make the surveillance more inefficient. To address the issue and automatically detect the violent scenes using surveillance cameras and Embedded GPU in real-time we have developed this project for the benefit of our society. As the alert is generated in real-time, the security can take action immediately to prevent any further damage or mishappening in the crowd. Our primary objective is to automatically differentiate between violent activities and non-violent activities through CCTV surveillance cameras and automatically display the security alert on the screen as soon as any violent activity is captured and thus ensuring the safety of our society.