Coombs, D., Herman, M., and Nashman, M. Real-time Obstacle Avoidance Using Central Flow Divergence and Peripheral Flow, presented at the International Conference on Computer Vision, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....and spherical fields of view [11] 12] 17] 24] Autonomous navigation, remote surveillance, and video conferencing are among the applications which should benefit from this technology. Already, wide angle imaging devices have begun to be incorporated into autonomous navigation systems [3] [15] 22] In this paper, we discuss the use of omnidirectional cameras for the recovery of observer motion (ego motion) an important problem in autonomous navigation. The ego motion problem can be stated as the recovery of observer rotation and direction of translation at a given instant of ....

D. Coombs, M. Herman, T. Hong and M. Nashman, "Real-time obstacle avoidance using central flow divergence and peripheral flow," ICCV '95, pp. 276-283.

....tasks. Since the enlarged field of view significantly improves the robustness of tracking or robot localization, wide angle images have been OMNIVIS 02 Workshop on Omni directional Vision, June, 2002, Copenhagen, Denmark (held with ECCV02) used in many applications of autonomous robot navigation [9, 10, 11] including motion estimation [12, 13] In the remaining of the paper, we will refer to planar perspective projection to indicate the pin hole camera model with a planar retina and designate the projection to a spherical retina as spherical projection. Gluckman and Nayar [14] showed that good ....

D. Coombs, M. Herman, T. Hong, and M. Nashman, "Real-time obstacle avoidance using central flow divergence and peripherical flow," Proc. of ICCV'95, 1995.

....set of behaviors which we have implemented in simulations and in real robots, and derive a purposive representation [1] of the environment in terms of a virtual corridor , based on which the robot reliably performs these navigational tasks. Our virtual corridor is an extension of Coombs et al. s [4, 3] conceptual corridor and of other minimalist representations designed for the basic navigational tasks considered here [15, 19] The inspiration for such representations comes from studies by Srinivasan [17] on the flight behavior of bees, which showed that bees perform seemingly complex ....

....comes from studies by Srinivasan [17] on the flight behavior of bees, which showed that bees perform seemingly complex flight maneuvers reliably based on very coarse information extracted from the optical flow field. These bee behaviors can be successfully used in groundbased robotic vehicles [4, 3, 15, 21]. In section 3 we present a short summary of the properties of the optical flow field during typical ground plane motion, and show how the information necessary for performing the navigational tasks can be derived from it. Since the robot moves, it is required to react in time to obstacles in its ....

D. Coombs, M. Herman, T. Hong, and M. Nashman. Real-time obstacle avoidance using central flow divergence and peripheral flow. IEEE Trans. on Robotics and Automation, 14(1):49--59, 1998.

....of obstacle avoidance exist in the literature. The problem of sensor based navigation has been previously examined by different approaches. Basically, these techniques can be distinguished according to the nature of the sensory input: stereo vision [4] monocular vision using the optical flow [2, 12], ultrasonic sensors [7, 13] or a laser range finder (LRF) like the SICK sensor which is actually mounted on our robot. In this paper, we present a method combining vision and a SICK sensor to perform a visual servoing task while avoiding obstacles. The basic idea is, in case of obstacles, to ....

D. Coombs et al. Real-time obstacle avoidance using central flow divergence and peripheral flow. In Proceedings of the International Conference in Computer Vision (ICCV), pages 276--283, Massachusetts, USA, June 1995.

....and numerically instable. Graphically, this instability results from the similarity of optic flow fields generated by rotation and lateral translation [15] The difference between those two field contributions is often hardly the magnitude of noise. Common approaches used by robotic researchers [3, 4, 5, 12] include the assumption of constraints to the motion pa rameters, for example to mere translation. Also qualitative detection without 3D reconstruction is proposed, avoiding the need of exact knowledge of the ego motion. The method for obstacle detection presented in section 3 exploits the ....

....of 25 ffi delivers obstacles from a minimal distance of 1:5 m to a maximal distance of 3:5 m in the central area and 8 m in the peripheral areas. In similar robotic systems, only qualitative scene descriptions are derived from optic flow, which allow only qualitative navigation approaches [2, 3, 4, 12]. Since the result of our obstacle detection approach is a local occupancy grid, standard robot navigation techniques can be implemented. ....

D. Coombs, M. Herman, T. Hong, and M. Nashman. Realtime obstacle avoidance using central flow divergence and peripheral flow. In Proc. 5th Int. Conf. on Computer Vision, pages 276--283, Cambridge, MA, USA, June 1995.

....and spherical fields of view [11] 12] 17] 24] Autonomous navigation, remote surveillance, and video conferencing are among the applications which should benefit from this technology. Already, wide angle imaging devices have begun to be incorporated into autonomous navigation systems [3] [15] 22] In this paper, we discuss the use of omnidirectional cameras for the recovery of observer motion (ego motion) an important problem in autonomous navigation. The ego motion problem can be stated as the recovery of observer rotation and direction of translation at a given instant of ....

D. Coombs, M. Herman, T. Hong and M. Nashman, "Real-time obstacle avoidance using central flow divergence and peripheral flow," ICCV '95, pp. 276-283.

....perceptual processing and allows the most flexible behavior. 1 Introduction In the last few years, a number of researchers have explored the use of optical flow for directly controlling a robot s behavior (Coombs Roberts, 1993; Duchon Warren, 1994; Sobey, 1994; Santos Victor et al. 1995; Coombs et al. 1995). This research has concentrated on methods for obstacle avoidance without modeling the world (Aloimonos, 1992) and has even produced robots which can dock (Santos Victor Sandini, 1994) and play tag (Duchon et al. 1995) These implementations on a robot platform have been inspired by theories ....

....signals that there is a wall in front of the agent. If is below a threshold, the agent will turn 180 ffi (the tau reflex ) 2. 3 Obstacle Avoidance For obstacle avoidance, we used the Balance (or Centering) Strategy which has proven useful in robot implementations (Duchon Warren, 1994; Coombs et al. 1995; Santos Victor et al. 1995) With the Balance Strategy, the agent moves so as to equate the average magnitude of flow detected on each side of the optical axis (which is tied to the heading direction both in simulation and in our robot) At each time step in the simulation, the instantaneous ....

Coombs, D., Herman, M., Hong, T., & Nashman, M. (1995). Real-time obstacle avoidance using central flow divergence and peripheral flow. In Prodeedings of the 5th International Conference on Computer Vision.

....both of which are essential for real time robotics. Another trend is to compute only normal flow [4] instead of the true optical flow. Although normal flow can be computed much more easily than full optical flow and may have sufficiently robust statistical properties for many useful tasks (e.g. [5, 6]) there will be many cases where full flow is preferable, if it can be computed efficiently and robustly. Nishihara [7] lists four criteria for a successful computer vision algorithm: noise tolerance, practical speed, competent performance, and simplicity. The first three correspond directly to ....

....real time display typically slows down the frame rate by about 25 . Example animations are available via the WWW at the URL listed in the author s address. Current work is being done to use this real time optical flow algorithm in place of the normal flow algorithm used for obstacle avoidance in [6]. The algorithm used in [6] did run in real time on an Aspex PIPE but only gave normal flow at edges, resulting in Figure 10: A more difficult sequence showing three frames of an approach to wooden chairs, with the magnitude plot. The chairs on the right show clearly despite their slow image ....

[Article contains additional citation context not shown here]

D. Coombs, M. Herman, T. Hong, M. Nashman, "RealTime Obstacle Avoidance Using Central Flow Divergence and Peripheral Flow", Proceedings of the Fifth International Conference on Computer Vision, Cambridge, MA, June 1995.

.... of an active vision system [12] Recent work involving autonomous mobile robot systems have used single image cues for obstacle detection and avoidance such as stereo disparity [6] optical flow [5] visual looming [15] peripheral optical flow [8] divergence of image flow and time to contact [7], and appearance based models of color and shape [21] Shigang et al. 23] have recently proposed a method for autonomous robot navigation along routes described by landmarks based on range and color information. The following sections describe our dynamic obstacle recognition and avoidance ....

D. Coombs, M. Herman, T. Hong, and M. Nashman. Realtime obstacle avoidance using central flow divergence and peripheral flow. In Proc. Fifth Inter. Conf. Computer Vision 156 166 180 190

....correlation. This replaces stereo previously implemented on the Datacube system which could operate at 15Hz, but the Datacube does not fit into an embedded system [13] Optical flow [4] can be implemented in similar fashion to the stereo on the VIP TIM, to support obstacle avoidance based on flow [5]. Sample Image Stereo Depth Image Figure 9: Results of the stereo algorithm Figure 9 presents an example of the results obtained by the stereo algorithm. The brighter shades of grey represent points in the scene that are closer to the robot. Likewise the darker shades of grey represent the points ....

David Coombs, Martin Herman, Tsai Hong, and Marilyn Nashman. Real-time obstacle avoidance using central flow divergence and peripheral flow. In Proc. 5th International Conference on Computer Vision, pages 276--283, June 1995.

.... on vision based skills similar to those I would like my robot to be endowed with or learn include road following systems that use template matching to stay in their lane [9] edge based segmentation for following corridors [7] and optical flow for collision avoidance and hallway centering [5, 11]. 3.2 Learning I have been studying what learning techniques have been applied to robotics and vision, but I m not yet sure which are best suited to the tasks I have in mind. The robot s behavior will not be learned from a tabula rasa every learned skill will be grounded in some ....

D. Coombs, M. Herman, T. Hong, and M. Nashman. Real-time obstacle avoidance using central flow divergence and peripheral flow. In ICCV, pp. 276-283, 1995.

No context found.

D. Coombs, M. Herman, T.-H. Hong, and M. Nashman. Real-time obstacle avoidance using central flow divergence and peripheral flow. IEEE Transactions on Robotics and Automation, 14(1):49--59, 1998.

No context found.

D. Coombs, M. Herman, T.-H. Hong, and M. Nashman. Real-time obstacle avoidance using central flow divergence and peripheral flow. In Proc. of ICCV 1995.

....5.5. Tracking We track the probe in 2 D as the arm moves towards its goal position using the predicted probe image velocity and sum of absolute differences (SAD) correlation algorithm [9] A predictive filter is used to filter and predict the probe position and velocity at the next time interval [12][13] 14] The prediction is based on a weighted sum of the current position and velocity and a history of past positions and velocities. Depending on the weights used, the predictions can be tuned to be more responsive to new readings or to respond smoothly over time. At each processing iteration, ....

Coombs, D. , Herman, M. , Hong, T. , Nashman, M. , "Real-Time Obstacle Avoidance Using Central Flow Divergence and Peripheral Flow", Proc. Fifth International Conference on Computer Vision, Cambridge, MA, June 1995.

....are combined by summing the egocentric heading changes dictated by both systems. It is possible for this strategy of combining behaviors to result in taking a heading that is dangerous because there is no way for behaviors to eliminate all dangerous headings from consideration. Coombs, et al. 11][12] also used flow to implement corridor following and used divergence to detect imminent collision. The present work achieves similar results using divergence alone and is, therefore, not limited to corridor following. The present system supports goal directed behavior while providing local ....

....goal directed behavior while providing local obstacle avoidance. The method of estimating optical flow described in this paper has been shown to detect obstacles as far away as 6 meters under good conditions. On the other hand, the flow returned from the PIPE image processing computer used in [11][12] was limited to a range of about 1 to 2.5 meters due to the difficulty of detecting edges of distant surfaces. The range of our system is even more remarkable given the coarse resolution (32x64) of the images used. In addition, our system implements both wide angle and narrow angle camera ....

[Article contains additional citation context not shown here]

D. Coombs, M. Herman, T.-H. Hong, and M. Nashman, "Real-time Obstacle Avoidance Using Central Flow Divergence and Peripheral Flow," IEEE Transactions on Robotics and Automation, 14(1):49--59, 1998.

....[30] on that bin which contains the midpoint of the histogram for that pixel s local window. In our case, however, the divergence data are already quantized to 8 bit values (necessary in ZT Z nm nm order to reduce the data bandwidth in the previous system s implementation [11]) so this extension is not necessary. Although the separable median filter is not guaranteed to find true median values, it reduces noise almost as well as a true median filter [26] By definition, the true median is greater than or equal to half the data in the data set and less than or equal to ....

....divergence estimate arising from an object rises reasonably smoothly as the object is approached, and the object continues to be visible until the robot is very near to it. These results represent a considerable improvement in both range and persistence of detection over the results reported in [11], in which objects were detected in the narrow range of 1 to 2.5 m from the vehicle. Based on these trials, an imminent collision function was derived for robot speeds up to 0.8 m s (Figure 18) 8.2 Gauntlettrials The robot ran a gauntlet of office chairs to demonstrate the system s ability to ....

[Article contains additional citation context not shown here]

D. Coombs, M. Herman, T.-H. Hong, and M. Nashman. "Real-time Obstacle Avoidance Using Central Flow Divergence and Peripheral Flow," In Proc. of ICCV 1995.

....basic navigation competence can free additional resources for attending to the environment. SAFE MOBILITY USING MOTION VISION ALONE The lure of using motion vision as a fundamental element in the perception of space drives this effort to use flow features as the sole cues for robot mobility [2]. Real time estimates of image flow and flow divergence provide the robot s sense of space. Building on the system described above, the robot steers down a conceptual corridor, comparing left and right peripheral flows. Large central flow divergence warns the robot of impending collisions at dead ....

David Coombs, Martin Herman, Tsai Hong, and Marilyn Nashman, "Realtime obstacle avoidance using central flow divergence and peripheral flow," In Proc. of ICCV 1995, the Fifth International Conference on Computer Vision, Cambridge, Massachusetts, June 1995.

.... algorithms hardware approach Category Type Difficulties Parallel computers Connection machine [7] 35] 48] 51] Parsytec transputer [42] and hybrid pyramidal vision machine (AIS4000 and CSA transputer) 13] high cost, weight and power consumption Special image processing hardware PIPE [1] [11] [39] 46] Datacube [34] and PRISM 3 [36] low precision Dedicated VLSI chips Vision Chips: gradient method [31] 44] correspondence method [12] 43] and biological receptive field design [14] 30] low resolution Non Vision Chips: analog neural networks [20] digital block matching technique ....

.... [20] digital block matching technique [4] 18] 49] coarsely quantized estimates Table 3: Real time motion estimation algorithms algorithmic approach Technique Algorithms Difficulties Sparse feature motion tracking [1] 28] 33] computing time to contact (and hence obstacle avoidance)[11] and segmentation [42] requirement of temporal filtering Special constraints constraint on motion velocity [9] constraint on projection [52] constraint on input images Efficient algorithm 1 D spatial search [3] separable filter design (Liu s algorithm[25] requirement of careful ....

[Article contains additional citation context not shown here]

Coombs, D., Herman M., Hong T. and Nashman, M., "Real-time Obstacle Avoidance Using Central Flow Divergence and Peripheral Flow", Proceedings of IEEE International Conference on Computer Vision, Cambridge, MA, 1995.

.... algorithms hardware approach Category Type Difficulties Parallel computers Connection machine [6] 26] 37] 39] Parsytec transputer [32] and hybrid pyramidal vision machine (AIS 4000 and CSA transputer) 11] high cost, weight and power consumption Image processing hardware PIPE [1] [9] [29] 35] Datacube [25] and PRISM 3 [27] low precision Dedicated VLSI chips Vision Chips: gradient method [23] 34] correspondence method [10] 33] and biological receptive field design [12] 22] low resolution Non Vision Chips: analog neural networks [17] digital block matching technique ....

....approach generally suffers from high cost and low precision. The most popular algorithmic method is to compute sparse feature motion. Recent advances in this approach have enabled versatile applications including tracking [1] 20] 24] computing time to contact (and hence obstacle avoidance)[9] and even segmentation [32] which were believed to be better handled with dense data. However, in order to interpret the scenes with only sparse features, these algorithms need to use extensive temporal information (e.g. recursive least squares, Kalman filtering) which is time consuming. ....

[Article contains additional citation context not shown here]

Coombs, D., Herman M., Hong T. and Nashman, M., "Real-time Obstacle Avoidance Using Central Flow Divergence and Peripheral Flow", Proceedings of IEEE International Conference on Computer Vision, Cambridge, MA, 1995.

....is measured in pixels. We track the probe as the arm moves towards its goal position using the predicted probe image velocity and sum of absolute differences (SAD) correlation algorithm. A predictive filter is used to filter and predict the probe position and velocity at the next time interval [7][15] 19] At each processing iteration, the search direction and window used for the SAD correlation is updated based on the predicted probe image position and velocity. In Figure 8, P is the computed 2 D probe position, Q is the predicted position, and V is the vector representing the probe ....

Coombs, D., Herman, M., Hong, T., Nashman, M., "Real-Time Obstacle Avoidance Using Central Flow Divergence and Peripheral Flow", ICCV, June 1995.

No context found.

Coombs, D., Herman, M., and Nashman, M. Real-time Obstacle Avoidance Using Central Flow Divergence and Peripheral Flow, presented at the International Conference on Computer Vision, 1995.

No context found.

Coombs, D., Herman, M., Hong, T.H. and Nashman, M.: Real-time Obstacle Avoidance Using Central Flow Divergence and Peripheral Flow. Int. Journal of Robotics and Automation. (1998) 14(1):49-59

No context found.

T. H. Hong D. Coombs, M. Herman and M. Nashman. Real-time obstacle avoidance using central flow divergence and peripheral flow. Int. Journal of Robotics and Automation, 14(1):49--59, 1998.

No context found.

David Coombs, Martin Herman, Tsai-Hong Hong, and Marilyn Nashman. Real-time obstacle avoidance using central flow divergence and peripheral flow. International Conference on Computer Vision, pages 276--283, 1995.

No context found.

D. Coombs, M. Herman, T. Hong and M. Nashman. Real-time obstacle avoidance using central flow divergence and peripheral flow. IEEE Transactions on Robotics and Automation, 14(1):49-- 59, February 1998.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC