Moving-object tracking and vision navigation based on selective attention mechanism

A novel algorithm, which is based on the Selective Attention Computational Model (SACM), is proposed in this paper for the target tracking and vision navigation task. In order to achieve a certain kind of auto-recognition goal for some real-time applications, a Fast Template Matching Optical Flow Estimation Technique (FTMOFET) and an Adaptive Region Growing and Merging Algorithm (ARGMA) are developed for motion detection and the calculation of the high-level semantic information for image region. At the same time, an information-classification-based selective attention model, which is implemented by the BP neural net and the fuzzy set theory, is utilized for the ascertainment of a moving target region. Unlike some traditional attention models, our SACM model is designed both for the aim of the target region selection and the distribution of the limited computational resources reasonably. Experiment results show this model can achieve a good performance for moving target recognition and tracking in most complex environment.