Visual self-localization of mobile robots is a fundamental problem in robot navigation. It is the premise of trajectory planning or vSLAM. In this paper, the research results of mobile robot vision self-localization in recent years are reviewed. The principle, advantages and disadvantages of various methods are introduced.
Research about robot vision is growing and more focusing on using image processing algorithm to enhance the sensing ability of a robot mobile to aware about its surroundings. This research using Kinect sensor which has 2 cameras, one RGB visible light and IR camera / depth camera both with 640×480 pixels at 30fps (frame per second). An object following method is applied on a 4WD differential mobile robot chassis equipped with Kinect sensor 1.0 running at 640×480 pixels, 30fps. The method using pixel coordinate averaging which summing all the pixel coordinate of the object of interest. Each calculation (vertical and horizontal) will result in a single integer value of coordinate in i or j showing the center of the object. Two tests object using symmetrical and non-symmetrical shows that the center object only moves slightly off about 1 pixel on j axis. The result of the calculation for symmetrical object on j axis is 9 (rounded down) and for the non-symmetrical object is 8 (rounded up). with the calculation on i axis is similar (on pixe114).
This paper provides an illumination robust direct monocular simultaneous localization and mapping (SLAM) method, which takes advantage of the RGB channel to enhance the lighting change invariance of the input frame series. By linearly combining the RGB channel, some illumination-insensitive components in the colour space are extracted and represented by three indicators in this paper. An optimization model is then provided to minimize the errors in these indicators using a Kalman filter (KF). These indicators serve to update the frames by keeping only the illumination-insensitive components. The illumination robust visual SLAM method based on these enhanced frames is thereby offered. In addition, the gradient magnitude is utilized to improve the distinctiveness of each pixel in the frame. Experiments on both artificial and natural datasets show that the provided method has a better illumination robustness than the state-of-the-art direct SLAM method.
—In this paper, the problem of finite-time stability for general non-Gaussian stochastic system with unknown state is investigated by using T-S fuzzy modeling. The objective is to control the probability density functions(PDFs) of the system output to follow a given PDFs. In order to describe the gray-box dynamics between PDFs of system output and controlled input, the well known fuzzy logic systems and the T-S fuzzy models are imported at the same time. Then a classical state observer is used to imitate the unknown state of the system. On account of the strong nonlinearity stochastic process, the square root B-spline is used to model the target PDFs. Finally, the favorable finite time stability can be achieved by employing convex optimization theory. Simulations study for a non-gaussian model are given to embody the superiority of the designed algorithm.
