In order to identify the social groups based on the pedestrian motion, the trajectories of pedestrians are required. Pedestrians have to detected and tracked to build their trajectories. The Fast HOG implementation of the Histogram of Oriented Gradients for Human detection and the Mixture of Gaussians based Backgroun substraction were used in this research.

A pedestrian detector with a three-level blob filtering approach has been developed to improve the pedestrian detection accuracy. The three-level blob filter uses pedestrian detections from Background Subtraction and Histogram of Oriented Gradients for Human Detection. The filters attempt to eliminate some of the non-pedestrian confusers found in real world situations. The false negatives in the Fast HOG detections which happen in the near camera region, false positives which arise due to illumination variations and the false positives which resemble the pedestrian body patterns are filtered out by the three levels of blob filtering. The three filters are used together with the Fast HOG and Background subtraction detections
to achieve high pedestrian detection accuracy.