Object Entity-based Global Localization in Indoor Environment with Stereo Camera

We present a new method for indoor environment global localization with object recognition and stereo camera. For environment modeling, we use 3D object position and depth information of the stereo camera. The 3D object position is computed with the depth of the local invariant features. Furthermore, only the depth information at horizontal centerline in image is used, where optical axis passes through. The depth information is similar to the data of the laser range finder. Therefore, we can build a hybrid local map which is composed of indoor environment metric map and object location map. Also, based on such nodes, sensing data and object recognition, we suggest a method for estimating the coarse pose and refined pose of a mobile robot. The coarse pose is obtained by means of object recognition and SVD based least-squares fitting. Based on the coarse pose, the refined pose is estimated with particle filtering algorithm. One contribution of this method is that it can avoid the local minima problem which might be occurred in a geometrically non-distinctive place with scan matching based global localization method. With basic real environment, we show the proposed method can be an effective vision-based global localization algorithm