The goal of this project is to understand selected human motion or gestures in surveillance video data.

Distinguish foreground objects

all moving objects can be considered as foreground objects, after background subtraction, an image that shows foreground objects clearly is wanted.

Detect people in those objects

human can be in any shape, like stand and talk, or sit in a chair, or move with a trolly, or maybe just stand in different direction that cant have an absolute human shape appear in the video, in this step the goal is to recognise all human as more accuracy as possible.

Understand unusual movements

after the human detection, focus on any unusual motion (which can be previously defined) of human, and count it as a special signal that dangerous enough for raising alarm.