Video surveillance installations have surged in recent years. However, up to 98 percent of CCTV footage remains unseen. Analytics and automation technologies deliver smart data to improve safety, increase efficiency and reduce costs required to analyse the footage required.
In the age of the pandemic, we require accurate, efficient, and easily accessible information about crowds in public spaces.
Pensees has come up with a solution to this need: Crowd counting, which counts the total number of people in moving crowds, and helps to provide powerful traffic flow measurement with real-time or periodic reporting. Using detection based, regression, hybrid and density-map based methods, the crowd count is predicted by using the density map for the input image then summing over the entire density map. The density map incorporates a novel loss function that is used here in our deep learning network. It can construct a density contribution probability model to improve its robustness at low cost of the computation. We do not use any external detectors or multi-scale architectures but it plays agreeably well against many state of the art approaches.
Crowd counting is important for applications like video surveillance and traffic control, and the technology developed here can be applied for both people crowd counting and vehicle traffic counting.