Several law enforcement agencies have deployed drone systems to monitor hostiles, keep an eye on foreign drug cartels, perform border control operations, etc. This paper acquaints a system for drone surveillance to identify violent people in public areas. Our model uses bottom-up approach for 2-D pose estimation. The model extracts 18 key-points (x-y coordinates) from the videos frame by frame. A deep neural network takes these key-points as features and classifies activities as violent/non-violent.