People Counting via Supervised Learning-Based 2D CNN-LR Model in Complex Crowd Images

People counting from images can be a natural action, but automated tracking of individuals and counting through a machine-learning model is a big challenge. For an intelligent city transportation system, emergency people planning, and a better congestion control system, it is necessary to have an effective crowd-counting method even if the crowd is dense and fast. The supervised pedestrian estimation techniques work with crowded images along with the labeled information. The existing people-counting techniques often failed to offer practical ways to include training models from labeled samples that result in lower accuracy. To overcome these shortcomings, a deep CNN-LR architectural model is addressed to present people in images efficiently. Mainly this framework incorporates a linear regression model with a deep convolution neural network (2DConvNet) having deep accumulated attributes. The challenging and benchmark Mall dataset is used to conduct the people-counting experiment and secured MAE and MSE are 1.65 and 2.23, respectively, which indeed obtained a state of art level performance than other real-time crowd counting mechanisms.