In this paper, we propose a general framework of a distributed vision system combining pedestrian features with a static map of a mobile robot. In indoor environment with complex architectural structures, mobile robots cannot find the optimal global path only with the static map except moving objects such as pedestrians outside the view of sensors. Therefore, we propose a new calibration approach about the transformation from the pedestrian features in distributed webcams to those in the static map. Robot Operating System (ROS) is used for the mobile robot static map building, global path planning, localization and navigation with the laser scanner. And webcams scattered in indoor environment are available for the distributed vision calibration and pedestrian detection. We assume that light conditions of all the webcams remain relatively constant in indoor environment, then multi-scale Histograms of Oriented Gradients (HOG) slid on normalized sub images accelerates the pedestrian detection, and we also test the detection using convolutional neural networks (CNN). Next, detected features are calculated with calibration parameters and added on the static map. Finally, a real-time global path is planned for our mobile robot given the dynamic map.