Abstract With the proliferation usage of video surveillance for safety, traffic control, and privacy purposes and with the constant growth of population, it is important to keep monitoring using Closed-Circuit Television (CCTV). With new upcoming developed technologies, new systems and algorithms are introduced and implemented to the crowd counting system today retrieving live video surveillance from the CCTV. However, recent studies show that there are some challenges still faced regarding the crowd counting system which uses the density estimation. The problems that occurred have resulted from the inaccuracy of the system that is caused by several factors. Factors such as the perspective distortion which is caused by the lack of data training and the method such as face detection is an ineffective method to determine the population density. Studies proposed have projected the idea of developing a more robust crowd counting methodology by implementing crowd counting by detection, clustering, and regression. Implementing these methods using the Convolutional Neural Network (CNN) will better the result of the detection since in CNN the image can be inputted and it will undergo several layers which will result in the system being able to differentiate one image from the other. With CNN the process of crowd counting will be able to be more advanced.
DOI: https://doi.org/10.1088/1755-1315/729/1/012029
Publish Year: 2021
