Visual Saliency Based Objects Recognition

The object recognition of mobile robot, a multidisciplinary complex technology, is based onimage processing, analysis and understanding. Nowadays, it has penetrated into the military, spaceexploration, medicine, industry, etc. As an important part of outdoor mobile robot’s autonomousnavigation and environmental recognition, it has received widespread attention from researchers.Inspired by biology vision system, visual selective attention mechanism simulates the biologyvision, and extract the image area information from a complex scene that easily draws observers’attention, which puts up a new idea for overcoming the computing of large data volume andimproving system’s real-time performance. In this paper, the problem of objects recognition basedon visual saliency is studied.Firstly, the research significance and background of scene object recognition based on visionare discussed. Then, the application status of machine vision, the development history of objectrecognition in natural scene, and the problems encountered in the study are introduced. On the basisof introducing the mechanism of human visual perception, we deeply research two main models ofthe mechanisms of selective attention: bottom-up selective attention and top-down selectiveattention, as well as the characteristics and applications of visual selective attention.Secondly, for the problem of visual significant area extraction, it can’t fully describe thecontent in the image only by using one feature to characterize the information of the image, so wepresents a visual significant area algorithm based on color and texture. On the Lab color model, thealgorithm uses Difference of Gaussian filter for image processing, uses a two-dimensional Gaborfilter to extract texture features of the image, and finally integrates the color and texture features toget the significant area of the image. The result of the experiment shows that this method caneffectively detect salient region, restrain non-significant area, and can be very good for post imageprocessing, such as object recognition, sign detection, robot self-localization and scene classifi-cation etc.Lastly, for the problem of object recognition, the method simulates biological visual function.First of all, extract the salient region of the image based on visual selective attention mechanismand segment the salient region from the background. Then, extract invariant SURF feature of thesalient region. In order to improve computing speed, K-means algorithm is used to cluster theextracted SURF features and forms visual words. Finally, classify objects by the Support VectorMachine. Experimental results show that the method used in this paper is effective and can solve the problem of object recognition, which is important for mobile robot to patrol in the public,execute military reconnaissance mission, explore under outer space environmental and so on.However, compared with the high recognition rate and efficiency of human vision, objectrecognition and scene understanding based on computer vision information still have manyproblems to be solved.