Nurul Arif Setiawan

Gaussian mixture model in improved HLS color space for human silhouette extraction

In this paper, we present an algorithm using Gaussian Mixture Model (GMM) for foreground segmentation which can differentiate shadow region from objects with simple criteria. In the algorithm, we have utilized the Improved HLS (IHLS) color space model as the fundamental description for image, instead of using raw RGB data. IHLS color space has an advantage over the standard RGB space to recognize shadow region from object by utilizing luminance and saturation-weighted hue information directly, without any calculation of chrominance and luminance. By exploiting this feature in GMM, we obtain adaptive background model with good sensitivity to color changes and shadow.

Real-time vision based gesture recognition for human-robot interaction

In this paper, we propose gesture recognition in multiple people environment. Our system is divided into two modules: Segmentation and Recognition. In segmentation part, we extract foreground area from input image, and we decide the closest person as a recognition subject. In recognition part, firstly we extract feature point of subjects both hands using contour based method and skin based method. Extracted points are tracked using Kalman filter. We use trajectories of both hands for recognizing gesture. In this paper, we use the simple queue matching method as a recognition method. We also apply our system as an animation system. Our method can select subject effectively and recognize gesture in multiple people environment. Therefore, proposed method can be used for real world application such as home appliance and humanoid robot.

Optical Flow in Dynamic Graph Cuts

In this paper, we propose an approach to calculate optical flow by using dynamic graph cuts framework. We use several good ingredients to estimate optical flow. We combine occlusion model for visual correspondence problem (V. Kolmogorov and R. Zabih, 2001) and illumination invariant optical flow (D. Freedman and M.W. Turek, 2005) to achieve robustness given large occlusion and illumination changes. And we make use the dynamic algorithm from (P. Kohli and P.H.S. Torr, 2005) to estimates dynamically changing MRF models of labeling problems in computer vision. The dynamic algorithm is shown faster than standard counterpart, thus promising real time application of graph cuts techniques to estimate optical flow.

Human-Robot Interaction using Context Awareness and Active Plane Model

In this paper, after deciding correctly intention of actor using probabilistic context awareness, we propose method to recognize gesture using active plane model. Firstly, this algorithm defines actors state in 5 states; [NULL], [OBJECT], [POSE], [Global Gesture], [Local Gesture]. Then, probabilistic context awareness decides state that has maximum probability value in actors present state. Using APM, This algorithm extracts shape information and distance information in disparity image that is input by stereo camera. Finally, APM recognizes sequence of gesture by one gesture using HMM, after extracting eigen-vector and distinguishing each gesture through PCA. In this paper, proposed algorithm shows that it can be utilized by HCI of intelligent robot through experiment

Gesture recognition based on context awareness for human-robot interaction

In this paper, we describe an algorithm which can naturally communicate with human and robot for Human-Robot Interaction by utilizing vision. We propose a state transition model using attentive features for gesture recognition. This method defines the recognition procedure as five different states; NULL, OBJECT, POSE, Local Gesture and Global Gesture. We first infer the situation of the system by estimating the transition of the state model and then apply different recognition algorithms according to the system state for robust recognition. And we propose Active Plane Model (APM) that can represent 3D and 2D information of gesture simultaneously. This method is constructing a gesture space by analyzing the statistical information of training images with PCA and the symbolized images are recognized with HMM as one of model gestures. Therefore, proposed algorithm can be used for real world application efficiently such as controlling intelligent home appliance and humanoid robot.

Multiple people gesture recognition for human-robot interaction

In this paper, we propose gesture recognition in multiple people environment. Our system is divided into two modules: Segmentation and Recognition. In segmentation part, we extract foreground area from input image, and we decide the closest person as a recognition subject. In recognition part, firstly we extract feature point of subjects both hands using contour based method and skin based method. Extracted points are tracked using Kalman filter. We use trajectories of both hands for recognizing gesture. In this paper, we use the simple queue matching method as a recognition method. We also apply our system as an animation system. Our method can select subject effectively and recognize gesture in multiple people environment. Therefore, proposed method can be used for real world application such as home appliance and humanoid robot.

Multiple people labeling and tracking using stereo for human computer interaction

In this paper, we propose a system for multiple people tracking using fragment based histogram matching. Appearance model is based on Improved HLS color histogram which can be calculated efficiently using integral histogram representation. Since the histograms will loss all spatial information, we define a fragment based region representation which retains spatial information, robust against occlusion and scale issue by using disparity information. Multiple people labeling is maintained by creating an online appearance representation for each person detected in the scene and calculating fragment vote map. Initialization is performed automatically from the background segmentation step.