Yeon-ho Kim

Improvement of dead reckoning accuracy of a mobile robot by slip detection and compensation using multiple model approach

Although dead reckoning based on odometry and inertial sensors is essential for a robotic localization system, none of previous works gives reliable and accurate position estimates on irregular terrain over long periods of time. Classical approaches use one estimator (such as a Kalman filter) with a single system model. However the single system model is not good to deal with both of slip and no-slip situations because of the dynamics changes. In this paper, a multiple model approach that uses two Kalman filters is presented: one Kalman filter accounting for no-slip condition and the other for slip condition. The Interacting Multiple Model (IMM) is adopted to switch two Kalman filters depending on whether slip occurs or not, and gives the weighted sum of two filter estimates. Experimental results are included to validate our approach.

Monocular SLAM with undelayed initialization for an indoor robot

This paper presents a new feature initialization method for monocular EKF SLAM (Extended Kalman Filter Simultaneous Localization and Mapping) which utilizes a 3D measurement model in the camera frame rather than 2D pixel coordinates in the image plane. The key idea is to regard a camera as a range and bearing sensor, of which the range information contains numerous uncertainties. 2D pixel coordinates of measurement are converted to 3D points in the camera frame with an assumed depth. The element of the measurement noise covariance corresponding to the depth of the feature is set to a very high value. And it is shown that the proposed measurement model has very little linearization error, which can be critical for the EKF performance. Furthermore, this paper proposes an EKF SLAM system that combines odometry, a low-cost gyro, and low frame rate (1-2 Hz) monocular vision. Low frame rate is crucial for reducing the price of the processor. This system combination is cost-effective enough to be commercialized for a real vacuum cleaning application. Simulations and experimental results show the efficacy of the proposed method with computational efficiency in indoor environments.

A method of remote control for home appliance using free hand gesture

We propose a method of robust and fully automated remote control for home appliance using free hand gesture. The performance of the proposed system is verified with a hardware embedded in an air conditioner.

Usinig Optical Flow as an Additional Constraint for Solving the Correspondence Problem in Binocular Stereopsis

A stereo matching algorithm for 3D structure reconstruction relies on the correlation of image features in a pair of images. Usually, two calibrated cameras are used to capture scenes, and the left and right images are rectified to reduce the search space from 2D image region to 1D line. Then, the disparity is defined by the horizontal difference between the corresponding points on the search line. To solve the correspondence problem, various image features such as image intensity, edge, color, infrared light, pixel motion, etc. are employed either separately or integrated together. Among these image features, pixel motion or optical flow has been rarely used as one of the matching features (A. Scheuing & H. Niemann, 1986, G. Sudhir et al, 1995, A. M. Waxman & J. H. Duncan, 1986). In this chapter, we focus on the use of optical flow as an additional constraint in solving the correspondence problem. The stereo matching algorithm for 3D structure reconstruction can be divided into two groups based on the matching primitives: intensity based matching (also referred to as area based matching) and feature based matching. The intensity based matching method searches the best matching points using only intensity values of pixels. The method can further be divided into two groups depending on the smoothness constraints in minimization of matching cost function: local (window-based) method and global method. More details on the intensity based method can be found in a survey by Scharstein (D. Scharstein & R. Szeliski. 2002, M. Z. Brown et al, 2003). The intensity based matching method produces a dense disparity map without any additional post-processing, but usually needs an exhaustive search. Since the intensity based method uses the intensity value at each pixel directly, this method may suffer from the varying illumination problem. The feature based matching method searches the best matching points using some special symbolic feature points, such as line, contour, corner, etc. Since this method calculates disparity values only on the pixels corresponding to the feature point, this method does not require an exhaustive search. Also, symbolic features are less sensitive to illumination changes than pixel intensity, so the calculated disparity is more reliable than that of the intensity based matching method in the case of varying illumination. However, this method O pe n A cc es s D at ab as e w w w .ite ch on lin e. co m

Detection and localization of people in living room using a new background subtraction method

We propose a new block-based background subtraction and a sophisticated background update algorithm for illumination invariance and reliable performance. The proposed method is employed and verified for an air conditioner using an embedded hardware.

Illumination Invariant Motion Estimation and Segmentation

Extracting moving objects from their background or partitioning them have been one of the most prerequisite tasks for various computer vision applications such as surveillance, tracking, human machine interface, etc. Though many previous approaches have been working in a certain level, still they are not robust under various unexpected situation such as large illumination change. In this paper, we propose a motion segmentation method based on our robust illumination invariant optical flow estimation. We present the superiority of our motion estimation method with synthesized images and improved segmentation results with real images.