R. Dutta and C. Weems. Parallel dense depth from motion on the image understanding architecture. Proceedings of the IEEE CVPR, 154--159, 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....flow algorithm which returns dense data even in areas of relatively low texture. This is the starting point of our implementation. a) Test mobile robot (b) Robot s view of the lab through its 115 degree wide angle lens Figure 2 Robot and Its View In correlation based flow such as in [5][17][21] the motion for the pixel at in one frame to a successive frame is defined to be the displacement of the patch of pixels centered at , out of possible displacements (where is an arbitrary parameter dependent on the maximum expected motion in the image) We determine the correct motion of the ....

R. Dutta. C. Weems, "Parallel Dense Depth from Motion on the Image Understanding Architecture", Proceedings of the IEEE CVPR, p.154-159, 1993

.... ( TWR94] Authors rarely report the computational time needed for their algorithms; when they do, it is on the order of many minutes per frame ( WM93] or require specialized hardware such as a Connection Machine ( BLP89a] WZ91] L88] Datacube ( N91] LK93] or custom image processors ([DW93]) One obvious reason calculating optical flow is so computationally intensive is that images are composed of thousands of pixels (which often motivates massively parallel implementations) Another reason is that optical flow is very sensitive to noise, and drastic steps (such as iterative ....

R. Dutta. C. Weens, Parallel Dense Depth from Motion on the Image Understanding Architecture, Proceedings of the IEEE CVPR, p.154-159, 1993

....in this paper including comparisons with other algorithms, and discusses other related issues such as subsampling and latency. Section 7 gives details of the real time software and hardware implementations of this algorithm. 2 Correlation based Optical Flow In correlation based flow such as in [8, 13, 14] the motion for the pixel at [x,y] in one frame to a successive frame is defined to be the determined motion of the patch P of pixels centered at [x,y] out of (2j 1) 2j 1) possible displacements (where j is an arbitrary parameter dependent on the maximum expected motion in the image) ....

....be calculated) optical flow measurement density is 100 . The independence of one pixel s chosen displacement from all other pixels displacements motivates massive parallel implementations such as the closeto real time implementation on the connection machine described in [8] Dutta and Weems [13] calculate structure from motion using the same basic shiftmatch winner take all algorithm implemented on the A B A C Constant time delay, variable distances. Constant distance, variable time delays. C B t 1 t Figure 1: As the maximum pixel shift increases linearly, the search area increases ....

R. Dutta, C. Weems, "Parallel Dense Depth from Motion on the Image Understanding Architecture", Proceedings of the IEEE 1993 CVPR, p.154-159, 1993

.... ( TWR94] Authors rarely report the computational time needed for their algorithms; when they do, it is on the order of many minutes per frame ( WM93] or require specialized hardware such as a Connection Machine ( BLP89a] WZ91] L88] Datacube ( N91] LK93] or custom image processors ([DW93]) One obvious reason calculating optical flow is so computationally intensive is that images are composed of thousands of pixels (which often motivates massively parallel implementations) Another reason is that optical flow is very sensitive to noise, and drastic steps (such as iterative ....

R. Dutta. C. Weens, Parallel Dense Depth from Motion on the Image Understanding Architecture, Proceedings of the IEEE CVPR, p.154-159, 1993

.... Authors rarely report the computational time needed for their algorithms; when they do, it is on the order of many minutes per frame ( WM93] or require specialized hardware such as a Connection Machine ( BLP89a] DN93] SU87] WZ91] L88] Datacube ( LK93] N91] custom image processors [DW93], or PIPE [KSL85] WWB88] ATYM] Techniques which can run in real time often impose strict restrictions on the environment; HB88] presents a technique that can segment in real time for tracking purposes, but requires that a textured object be moving in front of a relative textureless ....

....loose sense of being fast enough to be calculated on line and be usable in some reactive system; on the order of 4 5 frames per second at a minimum. It is possible to perform the bulk of the computations in customized silicon chips; one such (partial) implementation has been done in CMOS [C91] [DW93] calculates structurefrom motion using the same basic shift match winner take all algorithm implemented on the Image Understanding Architecture simulator; they report an estimated 0.54 seconds per frame using a maximum possible displacement j = 20. LK93] calculates optical flow (for the purposes ....

[Article contains additional citation context not shown here]

R. Dutta. C. Weems, Parallel Dense Depth from Motion on the Image Understanding Architecture, Proceedings of the IEEE CVPR, p.154-159, 1993

.... Authors rarely report the computational time needed for their algorithms; when they do, it is on the order of many minutes per frame ( WM93] or require specialized hardware such as a Connection Machine ( BLP89a] DN93] SU87] WZ91] L88] Datacube ( LK93] N91] custom image processors [DW93], or PIPE [KSL85] WWB88] Techniques which can run in real time often impose strict restrictions on the environment; HB88] presents a technique that can segment in real time for tracking purposes, but requires that a textured object be moving in front of a relative textureless back2 ground. We ....

....time delays, search over time is linear. calculated on line and be usable in some reactive system; on the order of 4 5 frames per second at a minimum. It is possible to perform the bulk of the computations in customized silicon chips; one such (partial) implementation has been done in CMOS [C91] [DW93] calculates structure from motion using the same basic shift match winner take all algorithm implemented on the Image Understanding Architecture simulator; they report an estimated 0.54 seconds per frame using a maximum possible displacement j = 20. LK93] calculates optical flow (for the purposes ....

R. Dutta. C. Weens, Parallel Dense Depth from Motion on the Image Understanding Architecture, Proceedings of the IEEE CVPR, p.154-159, 1993

....vision, it is desirable to find a method of calculating optical flow that is less sensitive to noise in the imaging process, gives a dense output independent of the structure in the image, and is computationally efficient. 2 Correlation based Optical Flow In correlation based flow such as in [3, 12, 15] the motion for the pixel at [x,y] in one frame to a successive frame is defined to be the determined motion of the patch P of pixels centered at [x,y] out of (2j 1) 2j 1) possible displacements. We determine the correct motion of the patch of pixels by simulating the motion of the patch ....

R. Dutta. C. Weems, Parallel Dense Depth from Motion on the Image Understanding Architecture, Proceedings of the IEEE CVPR, p.154-159, 1993

....and higher CM IUA ratio as usual, the influence of Coterie functions is distinct. It has also been observed that these ratios are more distinct the smaller the field is on which the Coterie function operates. 72 5. 4 The Depth Recovery Algorithm In this subsection, another vision algorithm ([3]) is briefly introduced and run on the 3 different machines. In this algorithm, a parallel dense depth map is created, with the input of temporally separated images from a forward moving sensor. Correspondences between them are established in parallel through correlation. The correspondences are ....

R. Dutta and C. C. Weems. "Parallel Dense Depth from Motion on the Image Understanding Architecture," , IEEE Computer Society Conference on Computer Vision and Pattern Recognition, New York City, New York, June 15--17, 1993, pages 154--159.

No context found.

R. Dutta, C. Weems, and E. Riseman, Parallel Dense Depth from Motion on the Image Understanding Architecture, Proc. 1993 DARPA IUW , Washington, DC. April 19-21, 1993.

No context found.

R. Dutta and C. Weems. Parallel dense depth from motion on the image understanding architecture. Proceedings of the IEEE CVPR, 154--159, 1993.

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