Empirical Study on Real Time People Counting using Deep Learning

In this digital era, many crowd-counting systems still rely on old-fashioned approaches like keeping registers, using people counters, and using sensors to count people at the door. These strategies are ineffective in situations where people's movement is fully random, extremely unpredictable, and dynamic. Video surveillance plays a key role in monitoring people in several places. Using videos from this video surveillance in algorithms like Median Flow,” Minimum Output Sum of Squared Error” (MOSSE), “Generic Object Tracking Using Regression Networks” (GOTURN), kernelized filters designed with deep learning can track the objects and count the people or customers and correlation filters like “Single Shot Detector” (SSD) to count the number of individuals entering and leaving a location. This paper gives a review of the above algorithms.