A Novel Frame Similarity Based Pedestrian Counting Approach in Surveillance Videos

Pedestrian detection and counting gains an important role in video surveillance for entrance/exit monitoring, customer behavior analysis, traffic management and public service management. This paper proposes a detection based people counting approach that can easily be applied to all real-life scenarios and provides much accurate result as compared with those of other existing methods. The proposed approach is divided into two phases. In the first phase, object detection is performed while in the second phase, detected frames are analyzed based on their features to provide the count of the persons in a given video. We evaluate our approach on PETS 2009, Mall, UCSD datasets. An additional dataset(Traffic dataset) is also used in order to verify the effectiveness of our model for multi-object counting. Experimental results on these datasets justifies the good performance of our proposed method for crowd counting.1