Visual navigation method based on BoW object recognition model

A new visual navigation method for mobile robot to navigate in dynamic environment is proposed. This method relies on some natural landmarks, which are common objects in indoor environment. Robot can establish recognition model and cognize various objects in the environment by monocular camera. Firstly, an image database of indoor common objects is established, and the SIFT (scale invariant feature transform) features are extracted from these images. Then, each image in the database is described as a bag of words (BoW) model. The linear SVM (support vector machine) is used to discriminate every class of images. Finally, An interactive hand-drawn-route-map is used to describe the dynamic indoor environment, where robot can get auxiliary route and natural landmark location and navigate successfully. The robustness of this method was verified by experiments in different aspects including the natural landmark change, destination area change and hand-drawn map deviation. A lot of experiments in the indoor environment show that the proposed method is very simple and highly effective, and has the advantage of high man-machine interaction and high adaptability to the dynamic environment.