Obstacle Avoidance And Positioning On Traveling Assistance For Visually Impaired People

According to statistics from the World Health Organization,there are approximately 217 million people in the world suffering from visual impairment,of which approximately 36 million are completely blind.Visually Impaired People(VIP)encounter many diffi-culties in their daily life,especially when traveling.Traditional travel aids mainly include white canes and guide dogs,but the detection range of white canes is limited and guide dogs are expensive,which cannot fully meet the needs of VIP for independent and safe travel.With the development of sensors and computer technology in recent years,smart devices have gradually moved from laboratories to daily life.More and more researchers are trying to use modern sensors and wearable devices to solve the problems encountered by VIP when traveling.Travel assistance for VIP mainly includes two aspects:assisting in avoiding obstacles and assisting in positioning.For obstacle avoidance,it not only needs to accurately identify the distribution of obstacles in front,but also to efficiently feed back the obstacle informa-tion to the user in an appropriate manner,which is often overlooked in most research on obstacle avoidance.In terms of positioning,although satellite positioning can meet the basic needs of VIP in outdoor environments,the positioning accuracy next to the building is low.It is difficult for VIP to determine whether the destination has been reached,that is,there is a last-meter problem.In addition,because satellite positioning cannot be used indoors,how to provide stable positioning assistance for VIP indoors is still a challenging topic.This thesis focuses on making up for the shortcomings in the above-mentioned travel aids research,and utilizes various sensors,such as RGB-D camera,panoramic annular lens and event camera,to provide them with reliable travel assistance.This thesis mainly in-cludes the following research contents.(1)Based on RGB-D cameras,this thesis proposes three independent obstacle avoid-ance methods to deliver three different levels of three-dimensional environmental informa-tion of depth images,obstacle distribution,and pathway information to VIP.In terms of interaction,this thesis follows the guidelines of sonification,and uses a variety of stereo sound effects to represent the detected environments,which improves the efficiency of information transmission.This thesis organizes twelve visually impaired volunteers to par-ticipate obstacle avoidance experiments to verify the reliability of the three methods.The characteristics of different methods are also analysed from multiple dimensions,such as environmental perception and obstacle avoidance.(2)Based on panoramic annular lens,an assisting positioning method for complex in-door scenes is proposed.When using the camera,the optical axis is vertically upward to shoot,and the information of the ceiling and part of the wall in the images is used to esti-mate the movement of VIP.This method firstly models the imaging method of panoramic annular lens,and then uses the visual odometry based on direct image alignment to cal-culate the movement of the camera,and at the same time restore the three-dimensional structure of the indoor scene.To assist the visually impaired locate in real buildings,this thesis also proposes a coordinate alignment method based on the QR code,which uses the position of the QR code in the building to project the motion information estimated by the visual odometry into the building.Experiments have proved that the panoramic positioning method proposed in this thesis is more robust than traditional camera-based methods,and is more suitable for assisting the visually impaired in indoor environments.(3)An outdoor positioning method based on an event camera is proposed,which uses the low latency and high dynamic range of event cameras to solve the last-meter problem in outdoor navigation for VIP.This method uses the high frame rate camera motion infor-mation provided by the inertial measurement unit to integrate the event stream data into event frames with low motion blur.Then,a sliding window-based non-linear optimization is used to fuse event frames,image frames and inertial sensor information to accurately es-timate the movement of the camera.This method also uses particle filter to fuse the result of camera motion estimation and the result of satellite positioning to obtain high frame rate positioning information in the geographic coordinate system,which is convenient to com-bine with existing navigation APPs.Experiment results show that the positioning accuracy of this method has reached the state-of-the-art level.With the help of event cameras,the robustness of the visual positioning algorithm to the rapid rotation of the human head is also significantly improved.From the perspective of stability and safety,this thesis conducts research on obstacle avoidance and positioning for VIP,and enhance the practicality of electronic travel aids.The proposed methods can not only be used in the field of travel aids for VIP,but also play an important role in the fields of robotics,autonomous driving,and augmented reality.