Human activity understanding is a branch of research in computer vision that has attracted a lot of attention for decades. Accurate identification of humans in video surveillance is fundamental prerequisite towards activities' understanding. Little or no research has been conducted for human detection in financial endpoint premises specifically Automatic Teller Machine (ATM) sceneries. The video surveillance settings have some unique features compared to others applications: static and non-uniform background, low resolution images, and lack of initial background model. The Histogram of oriented gradient technique was used to locate people in each frame of the surveillance video. Our framework achieved a precision of 88.71 and F-score of 56.41. © 2014 Springer-Verlag Berlin Heidelberg.