The Research On Transparent Obstacle Detection Technology For Visually Impaired People

There are about 285 million visually impaired people worldwide,about 39 million of whom are completely blind.Due to the lack of visual perception of the surrounding environment,they will encounter various inconveniences in daily life.Fortunately,with the continuous development of computer vision and sensor technology,there are many electronic travel aids(ETA)based on RGB-D sensor now.However,when these ETA encounter transparent obstacles,the RGB-D sensor is unable to calculate the correct depth because transparent objects have no fixed texture and image features.So these ETA cannot find transparent obstacles.Unfortunately,transparent objects are ubiquitous in our life,which means sometimes it is very dangerous for visually impaired people to use ETA.In order to resolve this problem,a transparent obstacle detection method based on multi-sensor fusion technology is proposed in this paper.By two ultrasonic sensors and a depth camera,the method can detect transparent obstacle in real time and output related navigation information.Firstly,the method need calibrate the corresponding detection area of ultrasonic sensors and the camera image.Then we calculate the depth value of the depth camera and ultrasonic sensors respectively in this area and judge whether there is transparent object in the corresponding detection area according to some discriminant conditions.After the transparent object is detected,the seed point growth algorithm is used to find the transparent object area in camera image and then the error depth in the original depth map is updated with the depth of ultrasonic sensors.When it inputs video stream data,a confidence parameter which can effectively avoid error detection,is introduced to ensure the continuity of the detection results in sequence image.In addition,the traversable area detection algorithm based on V-disparity is combined with the transparent obstacle detection algorithm proposed in this paper through the updated disparity map to form a traversable area detection system that can detect transparent obstacle.The system can plan traversable route in real time for visually impaired people and has better robustness than the original algorithm.The experimental results in actual scene show that the transparent obstacle detection algorithm proposed in this paper effectively solves the problem that the RGB-D sensor cannot detect transparent object and can detect common transparent obstacle and related information of position and distance.At the same time,the experimental results of quantitative data also prove that this algorithm has achieved good performance in real-time,accuracy and complex environment and it is able to meet the application requirements of visual assistance technology for the visually impaired.