A fast object segmentation method for mobile robots based on improved depth information

Due to the rapid development of mobile robots technology, the object recognition is of great practical significance. The real-time performance and robustness of object segmentation in cluttered environments is a considerable problem in robot vision. In this paper, a new object segmentation method using depth information is presented. Firstly, this approach obtains the object candidate region using the depth clue, then accomplished the depth filtering in the object candidate region. Next, the object region is extended to get the better edge information. Finally, the foreground is extracted and the segmentation results is realized on the color image. This method of object segmentation was tested on a real mobile robot platform and the results of experiments confirmed the excellent performance of the proposed method.

I. Introduction

Searching the environmental objects around the robot is a crucial precondition for the autonomous operation of mobile robot in indoor environments. Studies have shown that 83% of the human learning comes from the visual information [1]. In most cases, the object detection of mobile robot is realized by vision sensors. The visual data can be used for upper robotic tasks, including object recognition, location and mapping. In order to simplify these upper task, a segmentation is needed to eliminate the useless information in the scene. The primary goal of the segmentation for mobile robot is to obtain the object candidate region. This may be difficult to achieve, especially in most situation, the object is interlaced by cluttered background. Furthermore, during the movement of the robot, the background varies in the different occasions, hence the constant background model is difficult to define for mobile robots.