An Improvement of Positional Accuracy for View-Based Navigation Using SURF

Position and Rotation Estimation for Mobile Robots in Outside of Recording Path Using Ego‐Motion

“…Figures 12(a), 12(b), and 12(c) shows images of the experimental environment captured at the start position on the training and estimated paths in , the maximum value for the error in the position is 7 cm and 18 cm, respectively, and the average value, 4 cm and 5 cm. Even when the estimated path is on an angle with respect to the training path, the authors confirmed a position precision of a level similar to that of the conventional method [13]. Moreover, the estimation errors for the rotation in the estimated position indicated by the black triangles in Figs.…”

“…(a) and (b), the maximum value for the error in the position is 7 cm and 18 cm, respectively, and the average value, 4 cm and 5 cm. Even when the estimated path is on an angle with respect to the training path, the authors confirmed a position precision of a level similar to that of the conventional method .…”

“…Applied research in which the images are matched by using information on the robust changes in concentration (edges) for changes in the illumination in an environment has been reported on as a means to resolve the above problem. Moreover, the authors have also proposed a method to improve the estimation precision of the position while being robust with respect to changes in the environment traveled by the mobile robot by using speeded up robust features (SURF), which is robust with respect to changes in image illumination, scale, and rotation, for image referencing. In a paper, they tested through experiments using stationary and moving obstacles shading, one more form of change to the background color, and showed that this approach was robust up to a shading level of 55%.…”

“…Moreover, the authors have also proposed a method to improve the estimation precision of the position while being robust with respect to changes in the environment traveled by the mobile robot by using speeded up robust features (SURF), which is robust with respect to changes in image illumination, scale, and rotation, for image referencing. In a paper, they tested through experiments using stationary and moving obstacles shading, one more form of change to the background color, and showed that this approach was robust up to a shading level of 55%. However, all of these methods assume travel along a training path, and have the problem of position estimation becoming difficult when the robot deviates from a training path.…”

Development of a free viewpoint pedestrian recognition system using deep learning for multipurpose flying drone

“…Therefore, there is a need for a mechanism of human detection from bird's eye view, and a number of detection methods were proposed for this purpose. [6][7][8] Human detection models based on HoG features were proposed in the literature 6,7 ; however, HoG features are a histogram description of image intensity and direction, and are therefore sensitive to threedimensional (3D) appearance change of detected objects. In addition, human recognition was implemented via feature point matching using speed-up robust features (SURF), 8 but SURF points are also sensitive to 3D appearance change.…”