| With the increasing of global aging,the traditional electric wheelchair,as an important assistive product for the elderly,is not easy to control for some users.Developing a smart wheelchair with friendly interface and without changing the elderly’s daily habits is an inevitable trend under the background of intelligent pensions.An intelligent wheelchair called MIYABI Ⅳ with the functions of auto-following and natural gesture navigation was developed in this paper.The automatic following function was based on the fisheye camera and the detection and tracking fusion algorithm.The navigation was completed with simplified gestures,direction extraction algorithm and support vector machine classifier.Experiments showed that both modules have certain accuracy and robustness.The main research contents of this article are as follows:Gesture images were obtained for the navigation module from the depth camera.Then the gestures were extracted by skin detection and arm removal algorithms.A polygon fitting method was used to simplify the gestures,and a support vector machine model was trained for gesture recognition.The mean recognition accuracy was 99%.Human’s limbs were obtained for the auto-following module by a fisheye camera.Based on the correlation filtering and deep learning algorithms,a detection and tracking algorithms was proposed to track the user in 2D frames.Via lens calibration and coordinate conversion,the relative position of the user and the wheelchair could be calculated to assist in following human.Based on an electric wheelchair,MIYABI Ⅳ was modified by installing sensors and controllers.The control system of the wheelchair was developed based on PID.Experiments,including the gesture navigation and obstacle avoidance of the navigation module,and some experiments about the reliability of the auto-following module,were conducted.The experiments proved the algorithms and system developed in this paper were feasible and had potential in practical application. |
