Regresar

Feature weighting in competitive learning for multiple object tracking in video sequences

Abstract:

Object tracking in video sequences remains as one of the most challenging problems in computer vision. Object occlusion, sudden trajectory changes and other difficulties still wait for comprehensive solutions. Here we propose a feature weighting method which is able to discover the most relevant features for this task, and a competitive learning neural network which takes advantage of such information in order to produce consistent estimates of the trajectories of the objects. The feature weighting is done with the help of a genetic algorithm, and each unit of the neural network remembers its past history so that sudden movements are adequately accounted for. Computational experiments with real and artificial data demonstrate the performance of the proposed system when compared to the standard Kalman filter. © 2011 Springer-Verlag.