For robot visual Simultaneous Localization and Mapping (vSLAM), the first task is to initialize the image features. We present an improved feature initialization algorithm for robot SLAM, including detection of image features, selection of good features, calculation of image depth, and update of feature locations. Meanwhile, we extend the usage of the feature initialization algorithm to mapping and detection of features on a moving object. Experimentation is performed with real platform and the results show that the performance of the proposed feature initialization algorithm is efficient for visual SLAM and moving object detection.
