A neural-based crowd estimation system for surveillance in complex scenes at underground station platform is presented. Estimation is carried out by extracting a set of significant features from sequences of images. Those feature indexes are modeled by a neural network to estimate the crowd density. The learning phase is based on our proposed hybrid of the least-squares and global search algorithms which are capable of providing the global search characteristic and fast convergence speed. Promising experimental results are obtained in terms of accuracy and real-time response capability to alert operators automatically.

Describes a nonlinear preprocessing method to enhance the output performance of a network. The introduction of the nonlinear preprocessing method redistributes the distributions of input and output vectors, and makes the input and output variables more "orthogonal" that results in facilitating the network optimization. In some of the examples, this nonlinear preprocessing technique enables test set error to be reduced by a magnitude of 98%. Three applications of time-series predictions applied to evaluate the performance of the proposed method are presented.