Feature saliency based SLAM of mobile robot

We propose a stable CV-SLAM (Ceiling Vision based Simultaneous Localization and Mapping) technique, which uses both circle and corner features as landmarks in the scene and improves the process stability using saliency measurement. It provides a method which utilizes different feature detection algorithms to detect various key points. And then we measure saliency strength of every points to pick out more stable feature points and generate a hybrid map based on Delaunay triangles among these points. Moreover, we complete SLAM using an extended Kalman filter(EKF), which is fundament for robotic SLAM. Simulation results show the effects of proposed methods.