Bearing-only Simultaneous Localization and Mapping for Vision-Based Mobile Robots

Abstract

In this chapter, we proposed a vision-based approach to bearing-only SLAM in a 2dimensional space. We assumed the environment contained several visually distinguishable landmarks. This approach is inspired from techniques used in stereo vision and Structure From Motion. Our landmark initialization method relies solely on the bearing measurements from a single camera. This method does not require information from an odometer or a range sensor. All the object positions can be estimated in a landmark-based frame. The trade-off is that this method requires the robot to be able to move in a straight line for a short while to initialize the landmarks. The proposed method is particularly accurate and useful when the robot can guide itself in a straight line by visually locking on static objects. Since the method does not rely on odometry and range information, the induced map is up to a scale factor only. In our method, the distance || L1 - L 2 || of two landmarks is taken as the measurement unit of the map. The selection of L1 and L 2 is critical for the accuracy of