Attention-based Target Recognition Algorithm And Applications In The Mobile Robot

Target recognition for mobile robot has been a popular topic of study, which is thebasis for mobile robot to interact with the outside world. However，even a very simpletarget，it is very difficult to recognize it with a computer. On the other hand, humanitycan easily turn our attention to the target object and finish the target recognition taskfrom the complexity of the outside world. Therefore, it will be of great importance intheoretical and practical introducing human visual attention mechanism with powerfuldata filtering capabilities into the field of robot vision.This paper combines theoretical study and experimental analysis, and Expands withthe bottom-up and top-down visual selective attention mechanism modeling. The paperbriefly introduces the status of target recognition, and then analyzed and summarizedrecent physiology and psychology theory and attention model based on human visualattention mechanism, proposed a new attention-based target recognition algorithm.First the algorithm improves the existing model of attention mechanism. It integratesthe bottom-up and top-down attention mechanism by combination with the target colorfeature. It overcomes that the bottom-up attention mechanism is completely driven bythe image data, ignore the mission objectives, and lack of target guidance mechanism.Additional, it has a high computational efficiency and can meet real-time requirements.Secondly, the introduction of SIFT descriptor which has the property are translationalinvariance, rotational invariance and affine invariance to match the candidatestarget after extract them from last saliency map.Finally, apply the attention-based target recognition algorithm on kheperaII mobilerobot platform to obstacle avoidance and tracking in an unknown environment.Experimental results show that the algorithm can effectively recognize the target, and beable to meet real-time requirements for mobile robot.