M. Betke, N. Makris, Fast object recognition in noisy images using simulated annealing, in: Internat. Conf. on Computer Vision, 1994, pp. 523--530.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....optimization problem, and hence, unless the target region has some special structure, it is unlikely that the objective function is convex. Thus, in the absence of a good starting point, this problem will usually require some type of expensive search or global optimization procedure to solve [98]. In the case of visual tracking, the continuity of motion provides such a starting point. Let us assume that at some arbitrary time t t 0 the target region has the estimated configuration (t) We recast the tracking problem as one of determining a vector of o#sets, # such that (t #) t) # from ....

M. Betke and N. Makris, "Fast object recognition in noisy images using simulated annealing, " in Proceedings of the ICCV, pp. 523--530, 1995.

....and also as a Siemens Corporate Research Technical Report SCR94 TR 474, December 1993. The research in Chapter 4 is joint work with Nicholas C. Makris. An extended abstract of this work will be published in the Proceedings of the Fifth International Conference on Computer Vision in June 1995 [18]. It is also published as an MIT AI Memo 1510 and CBCL Memo 109 in December 1994 [17] Contents 1 Introduction 10 1.1 Contributions of the thesis : 12 1.1.1 Contributions to machine learning : 12 1.1.2 Contributions to ....

Margrit Betke and Nicholas C. Makris. Fast object recognition in noisy images using simulated annealing. In Proceedings of the Fifth International Conference on Computer Vision, Cambridge, Massachusetts, June 1995.

.... to the number of pixels in the background, and large motions can be performed (e.g. in the sequences that we tried, the face can move more than 20 pixels from frame to frame) In order to cope with such real conditions, we explore the use of stochastic methods such as Simulated Annealing (SA) [2], Genetic Algorithms (GA) 27,29] or Condensation (particle filtering) 6,17] for motion estimation. Although the techniques are very similar computationally speaking, here we make use of GA [29] within a coarse to fine strategy. Given the first image of the sequence, we manually initialize the ....

M. Betke and N. Makris. Fast object recognition in noisy images using simulated annealing. In International Conference Computer Vision, pages 523--530, 1994.

....the pixels of the candidate regions (i.e. the potential pictographs)# typically, regions are scaled to a fixed size and measures are taken to factor out lighting conditions. The resulting intensity features are subsequently fed into one of the established classifier tools: nearest neighbour (NNB) [2], radial basis functions (RBF) 10] polynom classifiers (PC) 12] and neural networks (NN) 1] 3] 12] or input to self designed schemes [5] Work by Kressel et al. 12] uses a combined classifier approach, employing a feed forward multi layer NN with spatial receptivefields for dimensionality ....

....a global approximator (PC) on large traffic sign databases. Not all work on traffic sign recognition requires a separate detection step as described in the previous section. Some work uses pictograph based classifiers directly in conjunction with a search algorithm. For example, Betke and Makris [2] use simulated annealing and Aoyagi and Asukara [1] use genetic algorithms. Because little or no (error prone) segmentation is required for these types of approaches, they have the potential to be very robust. In practice however, they tend to be quite slow and impractical for real time ....

M. Betke and N. Makris. Fast object recognition in noisy images using simulated annealing. In International Conference on Computer Vision, pages 523--530, 1995.

.... to find a good match if no particular architecture is used [5] Our search algorithm, based on a fast version of simulated annealing, estimates automatically the 4 parameters and avoids a suboptimal search strategy [10, 9] It uses a Metropolis dynamic and the temperature cooling is inverse linear [1]. Using deterministic refinement techniques, the affine or perspective transforms are estimated at each level and the matching procedure stops when the algorithm converges at the finest resolution. The total cpu time, on SUN SPARC20 workstation, is respectively 1.5s and 3s to estimate an affine ....

M. Betke and N.C. Makris. -- Fast object recognition in noisy images using simulated annealing. -- In ICCV95, pp.523--530, Boston, June 1995.

....chapter presents experimental results comparing di#erent image representations for object detection, in particular for detecting faces and people. Initial work on object detection used template matching approaches with a set of rigid templates or handcrafted parameterized curves, Betke Makris [ Betke and Makris, 1995 ] Yuille, et al. Yuille et al. 1992 ] These approaches are di#cult to ex81 tend to more complex objects such as people, since they involve a significant amount of prior information and domain knowledge. In recent research the detection problem has been solved using learning based ....

M. Betke and N. Makris. Fast object recognition in noisy images using simulated annealing. In Proceedings of the Fifth International Conference on Computer Vision, pages 523--20, 1995.

....attempts to use vision as the remote sensing mechanism responsible for detecting and identifying landmarks place us at the opposite extreme. While there are, at this point, systems which are capable of identifying at least some sets of landmarks in cluttered scenes fairly effectively (see [6, 7] for systems which work with a set of obstacles particularly chosen for certain robotic navigation applications, and and [42] for a wider survey) the automatic acquisition of a set of landmarks, 39 particularly when those landmarks occur in cluttered scene and must be segmented out before being ....

Margrit Betke and Nicholas C. Makris. Fast object recognition in noisy images using simulated annealing. In Proc. Fifth Intl. Conf. on Computer Vision, Cambridge MA, June 1995.

....the objects are difficult to model, there is significant variety in color and texture, and the backgrounds against which the objects lie are unconstrained. Initial work on object detection used template matching approaches with a set of rigid templates or handcrafted parameterized curves, [5, 71]. These approaches are difficult to extend to more complex objects such as people, since they involve a significant amount of prior information and domain knowledge. Other systems detect objects in video sequences focusing on using motion and 3D models or constraints to find people [60, 31, 25, ....

M. Betke and N. Makris. Fast object recognition in noisy images using simulated annealing. In Proceedings of the Fifth International Conference on Computer Vision, pages 523--20, 1995.

....be rewritten as I(m (t) t) I(0, t 0 ) and (2) becomes O(m) iI(m, t) I(0, t 0 )i 2 . 6) In general, 6) is a nonconvex objective function. Thus, in the absence of a good starting point, this problem will usually require some type of costly global optimization procedure to solve [35]. In the case of visual tracking, the continuity of motion provides such a starting point. Suppose that, at some arbitrary time t t 0 , the geometry of the target region is described by m(t) We recast the tracking problem as one of determining a vector of offsets, dm, such that m(t t) ....

 M. Betke and N. Makris, "Fast Object Recognition in Noisy Images Using Simulated Annealing," Proc. Int'l Conf. Computer Vision, pp. 523--530, 1995.

.... by a single template, these techniques typically use a set of templates to accommodate variations in pose, lighting and shape, by synthesizing views from a set of examples [5] and by using networks to learn an object class from several examples with varying pose and lighting [22, 17] Betke [4] uses simulated annealing for fast 2D object recognition (tra#c signs) in noisy images by matching a new scene to a set of templates generated by transforming model images. In deformable template matching [28, 6] templates are constructed to model the non rigid features of the object, and at ....

M. Betke and N. Makris "Fast Object recognition in Noisy Images using Simulated Annealing", Proc. ICCV, 523--530, 1995.

....paper presents experimental results comparing different image representations for object detection, in particular for detecting faces and people. Initial work on object detection used template matching approaches with a set of rigid templates or handcrafted parameterized curves, Betke Makris[1], Yuille, et al. 23] These approaches are difficult to extend to more complex objects such as people, since they involve a significant amount of prior information and domain knowledge. In recent research the detection problem has been solved using learning based techniques that are data driven. ....

M. Betke and N. Makris. Fast object recognition in noisy images using simulated annealing. In Proceedings of the Fifth International Conference on Computer Vision, pages 523--20, 1995.

.... drawn from an exponential distribution (meaning that small perturbations are more likely than larger ones) This optimisation technique is effectively a special case of simulated annealing [9] which has been successfully applied within the areas of object detection and recognition as reported in [8, 1]. To meet the real time requirements of the verification scenario, we employ a multiresolution scheme in the spatial domain. This is achieved by applying the combined gradient based and stochastic optimisation as described above to each level in a Gaussian pyramid. The estimate obtained on one ....

M. Betke and N. C. Makris. Fast object recognition in noisy images using simulated annealing. In ICCV'95, volume 1, pages 523--530, Washington, DC., 1995. Computer Society Press.

....there are several areas in the image having skin colour (for example, hands or unpainted wood) we use this step to determine which area is the face. Also, if our input image is a gray scale image, the colour discrimination is of no use. The approach described here is similar to the ones in [29] [37]. To find the face in the image (or a part of the image if colour discrimination has preceded this step) we define a five dimensional search space. The five parameters are x and y translation, rotation around the z axis, scaling, and x=y scaling (the relative face width) From a database of ....

....In each iteration, a new set of parameters is created by randomly perturbing the old set according to some temperature dependent distribution. The process is iterated a pre specified number of times, or until convergence has been reached. Use of simulated annealing can be found in [16] 29] [37]. In a step by step manner, the algorithm can be described as follows: 1. Assume a parameter space Omega and a parameter evaluation function f : Omega . The goal is to find the parameter set 2 Omega that maximizes f( 2. Define a stochastic parameter perturbation function G : Omega ....

M. Betke and N. C. Makris, Fast Object Recognition in Noisy Images Using Simulated Annealing, Artificial Intelligence Lab., MIT, Cambridge, MA, A.I. Memo No. 1510, 1994.

....intensities match the observed ones. Pearson s linear correlation coefficient r is often used for this type of assessment in engineering applications of object recognition. It has been shown that r is a very robust measure of the goodness of fit when the data are contaminated by noise (see e.g. Betke Makris 1995). By definition, the normalized correlation coefficient is r j = cov(t j ; d) p var(t j )var(d) 4) In this expression, cov represents the covariation operator, while var denotes variation. As before, d is the observational datacube and t j is one of our template models. Our numerous tests ....

Betke M., Makris N.C., 1995, Fast Object Recognition in Noisy Images Using Simulated Annealing. In: Proceedings of the Fifth international conference on computer vision. Cambridge, MA, June 1995. IEEE Computer Society Press. P.

....function defines how likely it is that an object with certain parameters is a car. The objective function combines evaluating the history of tracking a potential car with the correlation of the potential car with a deformable template of a car created on line using the method described in Ref. [1]. Various approaches for recognizing and or tracking cars have been suggested in the literature, for example, detecting symmetry points [6] approximating optical flow [9, 2] exploiting binocular stereopsis [3, 4] matching templates [4] and training a neural net [5] Related problems are ....

M. Betke and N.C. Makris. Fast object recognition in noisy images using simulated annealing. In ICCV, pages 523-530, Cambridge, MA, 1995. Also MIT-AI 1510.

....results to the classical computer vision problems of 3D motion and structure estimation for rigid objects. The Cramer Rao lower bound [ plays an essential role [19] in computer vision and has been widely used in the literature to address object motion and structure estimation problems [3, 4, 5, 6, 7, 8, 9, 13, 19, 20, 21, 22, 23, 25, 26, 27, 28]. Here we derive analytical expressions that axe necessary for the Cramer Rao lower bound to be a good approximation to the mean square error. Since the structure of real world objects can be approximated by a collection planar surfaces, we focus on the problem of estimating the pose and ....

M. Betke and N. C. Makris. Fast object recognition in noisy images using simulated annealing. In ICCV, pages 523-530, 1995.

....For the more general problem of recognizFigure 6: On the top: recognized stop sign. On the bottom: European no entry sign matches car roof. ing which object is in a scene image (i.e. not knowing the kind of traffic sign a priori) we ran 144 experiments with 18 scene images and 8 model images [3]. Table 1 contains some of the correlation values obtained in the experiments. For each scene image, our algorithm computes the highest correlation coefficient among the set of values obtained for each model (boldface values in Table 1) The model corresponding to the maximum correlation value is ....

M. Betke and N. C. Makris. Fast object recognition in noisy images using simulated annealing. MIT, AI Memo-1510, Dec. 1994.

....valuable for intelligent vehicles, which can use the recognition results to adjust their speeds or localize themselves in their environments [2] There are several conference papers on traffic sign recognition, for example, Refs. 1, 6, 11, 17, 19, 25] Our first results were published in Ref. [3]. Our method stands apart from previous approaches, because it is not restricted to edge detection as Refs. 1, 6, 17] and it does not rely on color information as Refs. 11, 19, 25] In principle, our approach could be extended by parameterizing color information. An overview of our traffic sign ....

M. Betke and N. C. Makris. Fast object recognition in noisy images using simulated annealing. In Proc. ICCV, pages 523--530, 1995. Also MIT AI Memo 1510.

No context found.

M. Betke, N. Makris, Fast object recognition in noisy images using simulated annealing, in: Internat. Conf. on Computer Vision, 1994, pp. 523--530.

No context found.

M. Betke and N. Makris, "Fast Object Recognition in Noisy Images Using Simulated Annealing." Proceedings of the Fifth International Conference on Computer Vision, pp. 523-530, June 1995.

No context found.

M. Betke, N. Makris, Fast object recognition in noisy images using simulated annealing, in: Internat. Conf. on Computer Vision, 1994, pp. 523--530.

No context found.

M. Betke and N. Makris, "Fast Object Recognition in Noisy Images Using Simulated Annealing," Proc. International Conference on Computer Vision, pp. 523-530, June 1995.

No context found.

Betke, M., Makris, N.: Fast object recognition in noisy images using simulated annealing. Technical Memo AIM-1510, Massachusetts Institute of Technology, Artificial Intelligence Laboratory (1995)

No context found.

# M. Betke and N. Makris, "Fast Object Recognition in Noisy Images Using Simulated Annealing," Proc. Int'l Conf. Computer Vision, pp. 523--530, 1995.

No context found.

M. Betke and N. Makris. Fast object recognition in noisy images using simulated annealing. In Proceedings of the Fifth International Conference on Computer Vision, pages 523--20, 1995.

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