Velocity field computation using neural networks

A new approach for optical flow (image velocity) fields computation is presented using computational neural networks. The computational procedure consists of three stages: estimation of the parameters of the neural network model, dynamic measurement of the perpendicular velocity components of the contours or region boundaries and computation of the image velocity fields. The parameters are estimated by comparing the energy function of the neural network with a constrained error function. The nonlinear velocity fields computation method is then carried out iteratively by using a dynamic algorithm to minimise the energy function simultaneously with the dynamic measurement of the perpendicular velocity components by a dynamic procedure. Experiments generate velocity fields that are meaningful and consistent with visual perception.