Abstract— The performance evaluation of an obstacle detection and segmentation algorithm for Automated Guided Vehicle (AGV) navigation using a 3D real-time range camera is the subject of this paper. Our approach has been tested successfully on British safety standard recommended object sizes and materials placed on the vehicle path. The segmented (mapped) obstacles are then verified using absolute measurements obtained using a relatively accurate 2D scanning laser rangefinder. Sensor mounting and sensor modulation issues will also be described through representative data sets. Index Terms—3D range camera, real-time, safety standard, ground truth, obstacle segmentation. I. INTRODUCTION bstacle detection and mapping are crucial for autonomous indoor driving. This is especially true for Automated Guided Vehicle (AGV) navigation in factory- like environments where safety of personnel and that of the AGV itself is of utmost importance. This paper describes the performance of an obstacle detection and segmentation algorithm using a 3D real-time range camera. The 3D range camera is based on the Time-Of-Flight (TOF) principle [8] and is capable of simultaneously producing intensity images and range information of targets in indoor environments. This range camera is extremely appealing for obstacle detection in industrial applications as it will be relatively inexpensive as compared to similar sensors and can deliver range and intensity images at a rate of 30 Hz with an active range of 7.5 m while incorporating no moving parts, such as a spinning mirror as in many off-the-shelf laser sensors. Since obstacle detection plays a critical role in autonomous driving, there has been much research on many different types of sensors, such as sonar [13], color/gray level cameras [2], FLIR (Forward Looking InfraRed) cameras [12], and stereo cameras [1, 6, 11, 14]. Most of the vision approaches are not applicable to indoor scenes due to lack of texture in the environment. Other researchers have proposed LADAR (Laser Detection And Ranging) sensors for detecting obstacles [3, 4, 5]. However, one-dimensional LADAR, which has been used in the AGV industry, is not suitable for the 3D world of factory environments and other complex volumes without moving the sensor during operation. Manuscript received November 16, 2004. R.V. Bostelman is with the National Institute of Standards and Technology, Gaithersburg, MD 20899 USA (phone: 301-975-3426; fax: 301- 921-6165; e-mail: roger.bostelman@nist.gov). T.H. Hong is with the National Institute of Standards and Technology, Gaithersburg, MD 20899 USA (e-mail: tsai.hong@nist.gov). R. Madhavan is with the National Institute of Standards and Technology, Gaithersburg, MD 20899 USA (e-mail: raj.madhavan@nist.gov). * Commercial equipment and materials are identified in this paper in order to adequately specify certain procedures. Such identification does not imply recommendation or endorsement by the National Institute of Standards and Technology, nor does it imply that the materials or equipment identified are necessarily the best available for the purpose. Our proposed approach to obstacle detection uses a low cost, 3D, real-time, range camera. First, we calibrate the camera with respect to the AGV so that we can convert the range values to 3D point clouds in the AGV coordinate frame. Second, we segment the objects which have high intensity and whose elevation values are above the floor of the operating environment on the AGV path. The segmented 3D points of the obstacles are then projected and accumulated into the floor surface-plane. The algorithm utilizes the intensity and 3D structure of range data from the camera and does not rely on the texture of the environment. The segmented (mapped) obstacles are verified using absolute measurements obtained using a relatively accurate 2D scanning laser rangefinder. Our approach has been tested successfully on approximate British safety standard recommended object sizes covered with cotton, cloth material and placed on the vehicle path. The AGV remained stationary as the measurements were collected for this paper. The American Society of Mechanical Engineers (ASME) B56.5-2004 standard [15] was recently changed 1 to allow non-contact safety sensors as opposed to contact sensors such as bumpers to be used on AGVs. Prior to the change, the B56.5 standard defined an AGV bumper as a “mechanically actuated device, which when depressed, causes the vehicle to stop.” With the current B56.5 standard change and with state- of-the-art non-contact safety sensors, vehicles can be shorter in length, excluding mechanical bumpers since these bumpers extend much farther in front and behind the vehicle than non- contact sensors. This in turn allows shorter vehicle turning radii and they can potentially move faster as objects can be detected well before the vehicle is close to an object. Ideally, the U.S. standard can be changed even further similar to the British EN1525 safety standard requirements [16]. Furthering the US safety standard will also provide support toward a unified, global safety standard for AGVs and other driverless vehicles. The paper has five sections: Section II describes the concept of obstacle detection and segmentation including the 1 not cited here since the change was not published prior to the date of this paper. Towards AGV Safety and Navigation Advancement - Obstacle Detection using a TOF Range Camera * R.V. Bostelman, T.H. Hong, and R. Madhavan O Proceedings of the 12th International Conference on Advanced Robotics (ICAR), Seattle, Washington, July 18-20, 2005.