Y. Raja, S. J. McKenna, S. G. Gong. Tracking and Segmenting People in Varying Lighting Conditions using Colour. Proc. 3rd Int'l Conf. On Automatic Face and Gesture Recognition, pp 228-233, 1998  Home/Search   Document Not in Database   Summary   Related Articles   Check

....light colors, e.g. under a sodium vapor lamp. There are numerous publications on hand segmentation. Two common methods are background subtraction and skin color segmentation. Obviously background subtraction is infeasible since there is no constant background. Color segmentation [3] 4] 5] [6] [7] 8] 9] 10] is more suitable in our case. Nevertheless, previous methods often use one static skin color model, which is inadequate for us. In the rest of this paper, we present a new way of segmenting hands with color information. 2.2. Problem Formulation We formulate the hand ....

Y. Raja, S. J. McKenna, S. G. Gong. Tracking and Segmenting People in Varying Lighting Conditions using Colour. Proc. 3rd Int'l Conf. On Automatic Face and Gesture Recognition, pp 228-233, 1998

....light colors, e.g. under a sodium vapor lamp. There are numerous publications on hand segmentation. Two common methods are background subtraction and skin color segmentation. Obviously background subtraction is infeasible since there is no constant background. Color segmentation [3] 4] 5] [6] [7] 8] 9] 10] is more suitable in our case. Nevertheless, previous methods often use one static skin color model, which is inadequate for us. In the rest of this paper, we present a new way of segmenting hands with color information. We formulate the hand segmentation problem as follows: The ....

Y. Raja, S. J. McKenna, S. G. Gong. Tracking and Segmenting People in Varying Lighting Conditions using Colour. Proc. 3rd Int'l Conf. On Automatic Face and Gesture Recognition, pp 228-233, 1998

....over time. Quantitative evaluation of the method was conducted on labeled ground truth video sequences taken from popular movies. 1 Introduction Locating and tracking patches of skin colored pixels through an image sequence is a tool used in many face recognition and gesture tracking systems [1, 4, 5, 7, 8, 9, 12, 13, 14]. An important challenge of any skin color tracking system is to accommodate varying illumination conditions that may occur within an image sequence. Some robustness may be achieved via the use of luminance invariant color spaces [13, 7] however, this method can withstand only changes that ....

....for each pixel. These features were used as input to an incremental EM algorithm that dynamically estimated the Gaussian mixture models for background and foreground. Pixels were grouped into skin color blobs. Kalman filters were used to filter spatial parameters for each blob. Raja, et al. [8] developed a tracking system that employed Gaussian mixtures to model skin, clothes and background. It was assumed that a skin color distribution can be modeled by a low order Gaussian mixture, where the number of components does not change over time or over a range of conditions. The system s use ....

Y. Raja, S.J. McKenna, and S. Gong. Tracking and segmenting people in varying lighting conditions using colour. In AFG, 1998.

.... point right hand to right pointing right wavea wave right hand above head urgent wave waveb wave right hand below head non urgent wave Gong at Queen Mary and Westfield College, London and Stephen McKenna at the University of Dundee, who are researching real time face detection and tracking [18, 20, 19, 27]. The standard RBF and TD RBF networks have already been shown to work well with such image sequences for face recognition tasks [14, 15] We are specifically interested in the areas of motion within each image, so each frame is differenced with the previous one: any pixel in the current frame ....

Y. Raja, S. J. McKenna, and S. Gong. Tracking and segmenting people in varying lighting conditions using colour. In Proceedings of IEEE International Conference on Automatic Face & Gesture Recognition, pages 228--233, Nara, Japan, 1998. IEEE Computer Society Press.

....roughly 5 seconds) for a total of 2832 images. These image sequences are the result of our collaboration in the ISCANIT project with Shaogang Gong at Queen Mary and Westfield College, London and Stephen McKenna at the University of Dundee, who are researching real time face detection and tracking [18 20, 28]. The standard RBF and TD RBF networks have already been shown to work well with such image sequences for face recognition tasks [14, 15] We are specifically interested in the areas of motion within each image, so each frame is differenced with the previous one: any pixel in the current frame ....

Y. Raja, S. J. McKenna, and S. Gong. Tracking and segmenting people in varying lighting conditions using colour. In Proceedings of IEEE International Conference on Automatic Face & Gesture Recognition, pages 228--233, Nara, Japan, 1998. IEEE Computer Society Press.

....over time. Quantitative evaluation of the method was conducted on labeled ground truth video sequences taken from popular movies. 1 Introduction Locating and tracking patches of skin colored pixels through an image sequence is a tool used in many face recognition and gesture tracking systems [1, 4, 5, 7, 8, 9, 12, 13, 14]. An important challenge of any skin color tracking system is to accommodate varying illumination conditions that may occur within an image sequence. Some robustness may be achieved via the use of luminance invariant color spaces [13, 7] however, this method can withstand only changes that ....

....for each pixel. These features were used as input to an incremental EM algorithm that dynamically estimated the Gaussian mixture models for background and foreground. Pixels were grouped into skin color blobs. Kalman filters were used to filter spatial parameters for each blob. Raja, et al. [8] developed a tracking system that employed Gaussian mixtures to model skin, clothes and background. It was assumed that a skin color distribution can be modeled by a low order Gaussian mixture, where the number of components does not change over time or over a range of conditions. The system s use ....

Y. Raja, S.J. McKenna, and S. Gong. Tracking and segmenting people in varying lighting conditions using colour. In AFG, 1998.

....over time. Quantitative evaluation of the method was conducted on labeled ground truth video sequences taken from popular movies. 1 Introduction Locating and tracking patches of skin colored pixels through an image sequence is a tool used in many face recognition and gesture tracking systems [1, 4, 5, 7, 8, 9, 12, 13, 14]. An important challenge of any skin color tracking system is to accommodate varying illumination conditions that may occur within an image sequence. Some robustness may be achieved via the use of luminance invariant color spaces [13, 7] however, this method can withstand only changes that ....

....features were used as input to an incremental Expectation Maximization Algorithm (EM) that dynamically estimated the Gaussian mixture models for background and foreground. Pixels were grouped into skincolor blobs. Kalman filters were used to filter spatial parameters for each blob. Raja, et al. [8] developed a tracking system that employed Gaussian mixtures to model skin, clothes and background. It was assumed that a skin color distribution can be modeled by a low order Gaussian mixture, where the number of components does not change over time or over a range of conditions. The system s use ....

Y. Raja, S.J. McKenna, and S. Gong. Tracking and segmenting people in varying lighting conditions using colour. In AFG, 1998.

....is able to independently perform the evaluation of range sensor data and the path planning in an appropriate processing frequency. 2 Vision Based Person Detection Earlier person recognition systems were often based on one single cue at the detection level, mainly motion [11] 13] or color [4] [10]. However, the claim for systems working in natural and complex environments requires a higher level of robustness and independence from constraints with respect to the application scenario. To comply with these requirements, many systems use a combination of multiple cues to segment a person from ....

Y. Raja, S. J. McKenna, and S. Gong, "Tracking and Segmenting People in Varying Lighting Conditions," in Proc. 3rd Int. Conf. on Automatic Face and Gesture Recognition, pp. 228-233, 1998.

....UV luminance chrominance space to detect the color of skin pixels. This has the advantage of avoiding any color transformation since the Y UV color space is also used in the MPEG standard. The skin detection works by modeling the skin colors using a Gaussian mixture distribution similar to that of [53], and then segmenting the image with a multiscale Bayesian segmentation algorithm known as sequential maximum a posteriori (SMAP) 54] The Gaussian mixture density models skin pixels with a multimodal distribution. This is important since it allows for a wide variety of possible skin colors or ....

Stephen McKenna Yogesh Raha and Shaogang Gong, "Tracking and segmenting people in varying lighting conditions using colour," in Proceedings of International Conference on Automatic Face and Gesture Recognition, Nara, Japan, April 1998.

....used to drive our body tracker are skin colour, image motion and coarse intensity information, namely hand orientation. Pixel wise skin colour probability has been previously shown to be a robust and inexpensive visual cue for identi cation and tracking of people under varying lighting conditions [22]. Skin colour probabilities can be computed for an image and thresholded to obtain a binary skin image, an example is shown in Figure 2(b) Here image motion is naively computed as the thresholded di erence between pixel intensities in successive frames; an example is shown in Figure 2(c) Skin ....

Y. Raja, S. J. McKenna, and S. Gong. Tracking and segmenting people in varying lighting conditions using colour. In Proc. 3rd IEEE Int. Conf. on Automatic Face and Gesture Recognition, pages 228-233, Nara, Japan, 1998.

....as a platform for investigating visually mediated interaction (VMI) methodologies. The current system uses a single pan tilt zoom camera as its only input, and integrates the following perceptual modules: 1) pixel wise motion from frame differencing, 2) pixel wise skin colour classification [8], 3) clustering into potential regions of interest, 4) support vector machine (SVM) for face detection [5] 5) person tracker to track head and hands, 6) gesture recognition [7] and (7) head pose estimation using similarity to prototypes [9] In order to operate in real time, it is essential ....

....The first step of bootstrapping is the calculation of a skin colour probability for each pixel in the image. Probabilities come from a mixture of Gaussian probability model in hue saturation space. The model is trained beforehand from example pixels using the Expectation Maximisation algorithm [8]. Both a foreground and a background model are used and combined using Bayes rule. The image is subsampled in this step to reduce computational expense. The skin colour probabilities are thresholded to give a binary image of skin non skin pixels. An example of this output is shown in Figure 1. ....

[Article contains additional citation context not shown here]

Y. Raja, S. J. McKenna, and S. Gong. Tracking and segmenting people in varying lighting conditions using colour. In Int. Conf. on Auto. Face and Gest. Recog., pages 228--233, Nara, Japan, 1998.

....rotations in depth result in non linear transformations. Face detection is needed in order to bootstrap the tracking and pose estimation process. Methods for face detection usually assume frontal or near frontal views and tend to be computationally expensive. They have been based upon colour [13], silhouette, spatial configuration of facial features and pattern recognition techniques (see [14] for references) Face detection is not the main topic of this paper and the systems described here used an appearance based matching scheme for detection and tracking. Further references on methods ....

Y. Raja, S. J. McKenna, and S Gong, "Tracking and segmenting people in varying lighting conditions using colour," in 3rd International Conference on Face and Gesture Recognition, Nara, Japan, 1998.

....framework used for estimating the parameters P = x; y; x ; y ) by matching the head models to the visual input over time [3] 4. 1 Model Matching The process can be bootstrapped using both motion and colour cues, a frontal view face detector, blink detection or some combination of these cues [7 9]. The process then proceeds as follows. In each frame, a region of the input image is searched by scanning the head model over it in order to nd the image location and head pose which provide the best match. This can be thought of as a search in a 4 D parameter space for the parameter values ....

Y. Raja, S. McKenna, and S. Gong, \Tracking and segmenting people in varying lighting conditions using colour," in IEEE Int. Conf. on Automatic Face and Gesture Recognition, Nara, Japan, April 1998.

....British Machine Vision Conference 4 of both face position and pose. The tracker is made more robust by incorporating skin colour information to determine the approximate face position. In face, skin colour is an ideal cue because it can be inexpensively computed in absolute terms for each frame [7]. Using skin colour, a separate head tracker supplies the bounding box of the head in the image. While the head box is generally larger than the face box, the displacement between head and face box position is used to constrain the tracker. In particular, correlations between head pose and the ....

Y. Raja, S. J. McKenna, and S. Gong. Tracking and segmenting people in varying lighting conditions using colour. In Proc. 3rd IEEE Int. Conf. on Automatic Face and Gesture Recognition, pages 228-233, Nara, Japan, 1998.

....track of both face position and pose. The tracker can be made more robust by incorporating additional visual information such as skin colour to determine the approximate face position. Skin colour is an inexpensive but effective visual cue that can be easily computed in real time at each frame [8]. Using skin colour, a separate head tracker can be used to supply a bounding box of the head position in the image. While the head box is generally larger than the face box, the displacement between head and face box position provides an additional constraint. In particular, correlations between ....

Y. Raja, S. J. McKenna, and S. Gong. Tracking and segmenting people in varying lighting conditions using colour. In Proc. 3rd IEEE Int. Conf. on Automatic Face and Gesture Recognition, pages 228--233, Nara, Japan, 1998.

....feature points on different faces are required and as a result, real time performance is obtained. The models are constructed from aligned image data labelled with pose angles. Efficient focus of attention based on colour and motion cues is used to bootstrap face image search in the image plane [6, 7]. 2 Acquisition of Labelled Views across the View sphere In order to build appearance models, example views labelled with 3D pose angles (both tilt and yaw) are required. A system was designed that utilises both a magnetic sensor attached to the subject s head and a camera calibrated relative to ....

Y. Raja, S. McKenna, and S. Gong, "Tracking and segmenting people in varying lighting conditions using colour," in IEEE International Conference on Automatic Face and Gesture Recognition, Nara, Japan, April 1998.

....tracking, therefore avoiding the need to perform online feature detection and correspondence which are computationally both expensive and problematic. 3 Bootstrapping Colour based segmentation can provide robust and very fast focus of attention for the initialization of the affine parameters [8]. Here we adopt multi colour Gaussian mixture models to perform real time object detection and focus of attention. The mixture models were estimated in twodimensional hue saturation colour space. Such representations are chosen to permit some level of robustness against brightness change. ....

....colour space. Such representations are chosen to permit some level of robustness against brightness change. Probabilities are computed for pixels in an image search space and the size and position of the object are estimated from the resulting probability distribution in the image plane [8]. An example of colour based, real time, coarse segmentation using a mixture of four Gaussians can be seen in Figure 1. 3.1 Adaptive Attentional Window The most computationally expensive operation in recovering affine parameters is to recursively warp the image relative to its center in order to ....

Y. Raja, S. McKenna, and S. Gong, "Tracking and segmenting people in varying lighting conditions using colour," in FG '98 (These Proceedings), 1998.

No context found.

Y. Raja, J. McKenna, and S. Gong. Tracking and segmenting people in varying lighting conditions using color. In Int. Conf. on Automatic Face- and Gesture-Recognition, pages 228--233, Nara, Japan, April 14-16, 1998.

No context found.

Y. Raja, J. McKenna, and S. Gong. Tracking and segmenting people in varying lighting conditions using color. In Int. Conf. on Automatic Face- and Gesture-Recognition, pages 228-233, Nara, Japan, April 14-16, 1998.

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