M. R. M. Jenkin and A. Jepson. Detecting floor anomalies. ARK Tech. Rep. in preparation.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....the number of layers in a distribution using Bayesian evidence and MDL Thomas F. El Maraghi Department of Computer Science University of Toronto March 1998 Copyright Thomas F. El Maraghi, 1998 1 1. Introduction Layered models have become a popular tool in the field of computer vision [1] 4][7][8] 9] 10] 17] 18] 19] Sensory data, such as the output of a CCD, is typically the result of multiple real world processes, such as motion boundaries, occlusion, and texture variation. In order to interpret such data accurately, it is necessary to use sufficiently rich and yet manageable models. ....

....boundaries, occlusion, and texture variation. In order to interpret such data accurately, it is necessary to use sufficiently rich and yet manageable models. Gaussian mixture models, which allow a distribution to be described as an ensemble of simple density functions, provide an elegant solution [7][8] 9] 10] 13] Mixture models can be fit to a data set using the expectation maximization (EM) algorithm. EM has been used by many authors, including [4] 7] 8] 9] 10] 12] 17] 18] 19] Its popularity arises from two extremely useful properties. First, it is guaranteed that successive iterations of ....

[Article contains additional citation context not shown here]

M. Jenkin and A. Jepson, Detecting floor anomalies, E Hancock, editor, Proceedings of the British Machine Vision Conference, BMVC-94, pp. 731-740, UK, 1994.

....a vision based approach to obstacle detection. Most approaches to visual obstacle detection exploit motion cues for locating obstacles. Furthermore, an assumption that is often made is that vehicle motion is confined to a surface that is either planar or can be approximated locally by planes [3, 2, 8, 11, 4, 14]. The existence of a planar ground gives rise to a phenomenon known as motion parallax in the psychophysics literature [5] A moving observer, perceives objects extending vertically from the ground to move differently from their immediate background. Various techniques for obstacle detection based ....

....valid and should be avoided when possible. Carlsson and Eklundh [2] assume a camera with unrestricted motion and predict the egomotion and the equation of the ground plane from long image sequences. Obstacles are identified in regions whose motion differs from that predicted. Jenkin and Jepson [8] apply the EM algorithm to obtain maximum likelihood estimates of the parameters of a mixture model describing the disparity field computed with phase based techniques from a calibrated stereo pair. The probability that a point does not belong to the floor is then computed from the ownership ....

M.R.M. Jenkin and A. Jepson. Detecting Floor Anomalies. In Proceedings of BMVC'94, pages 731--740, 1994.

No context found.

M. R. M. Jenkin and A. Jepson. Detecting floor anomalies. ARK Tech. Rep. in preparation.

....converted to 8 bit grey scale images used by the tracker. As described in x2.1, we model the scene as a set of two dimensional convex polygons. To obtain estimates for the object motions we use a view based tracking algorithm similar to the optical flow and stereo disparity algorithms described in [12, 11]. The input to the tracker consists of the image sequence, a set of object template images (including a polygonal outline for each object) and an estimate for the object positions in the first frame of the sequence. In addition, we provide an estimate for the position of the table top which is ....

Michael Jenkin and Allan D. Jepson. Detecting floor anomalies. In Proceedings of the British Machine Vision Conference, pages 731--740, York, UK, 1994.

....describe the various components of our implementation below. 4. 1 Configuration To acquire the position and orientation of the object polygons for each frame we use a viewbased tracking algorithm similar to the optical flow and stereo disparity algorithms described by Jepson and Black (1993) and Jenkin and Jepson (1994). In particular, a template image is provided for each of the objects, along with a polygonal outline describing the object shape within the template. Given an initial guess for the positions of the objects in the first frame, the tracking algorithm then estimates the two dimensional position and ....

Jenkin, M., & Jepson, A. D. (1994). Detecting Floor Anomalies. In Proceedings of the British Machine Vision Conference, pp. 731--740, York, UK.

....4 Optic flow was recovered from 4 The robot s speed was not measured, but was the equivalent of a fast walk. British Machine Vision Conference the sequence using a method that fits flow in image regions (patches) to functions that are either affine or rational in image coordinates x [11]. The were recovered by considering patches in a pair wise manner 5 : 6 flow samples were generated for each patch, using the 4 corners of each patch plus two interior points. The constraints were clustered according to the method in Section 4.1 and gave estimates for two translational ....

Jenkin M, Jepson A. Detecting Floor Anomalies. Proceedings of the British Machine Vision Conference, 1994

....to build the maximum floor coverage. The verification process uses the distance from the reference floor plane to accept or reject the region. 5.1. 2 Floor anomaly detection using stereo vision Another approach is to use stereo vision to verify that the floor in front of the robot is solid[10]. In a typical stereo vision application, objects in one camera are matched with objects in the other and these correspondences coupled with the known geometry can be used to identify the three dimensional location of structure in the environment. Perhaps the most difficult task in stereopsis is ....

M. R. M. Jenkin and A. Jepson. Detecting floor anomalies. In Proc. British Machine Vision Conference, pages 731--740, 1994.

....of the 3D coordinate frame. Then, for pinhole camera models, Faugeras [3] has shown that the mapping from a point, w r , in the right image to the corresponding point, w l , in the left image is given by w l = K( n; d) w r ; 2) where K( n; d) is a 3 Theta 3 matrix. We have shown [9] that the same form of equation suffices for the more general camera models described in the previous section, and we have derived a closed form expression for K in terms of L, R, n and d. It turns out that K( n; d) is a linear function of ( n; d) That is, using the slightly unusual notation ....

.... n Gammad : 4) Here M(L;R) is a 9 Theta 4 matrix which depends only on the left and right calibration matrices L and R. This makes it a relatively simple matter, using least squares, to convert a coefficient vector k to the corresponding parameters n, d for the ground plane in 3D [9]. Note that we can rewrite equation (2) in terms of the more familiar image coordinates x l and x r as x l 1 = m 1 ( x r ; k) k 1 x r 1 k 2 x r 2 k 3 ) k 7 x r 1 k 8 x r 2 k 9 ) x l 2 = m 2 ( x r ; k) k 4 x r 1 k 5 x r 2 k 6 ) k 7 x r 1 k 8 ....

[Article contains additional citation context not shown here]

M. R. M. Jenkin and A. Jepson. Detecting floor anomalies. ARK Tech. Rep. in preparation.

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