Research On Visual Location And Perception Of Unmanned Vehicle In Indoor Passageway

With the development of computer vision,pattern recognition and AI technology,the use of unmanned vehicle is increasing.In the past,unmanned vehicles are mainly used in specific scenes such as factories.In order to use unmanned vehicles in more indoor scenes,more general and efficient navigation/location algorithms need to be studied.In the indoor passageway,there are few feature points It is hard to location and nav-igation in such environment.The unmanned car may deviate from straight line driving because of the error in mechanical and motor.A real-time deviation detection algorithm is needed.It is a common mission for unmanned vehicle to move to the front of a door and through a door.This thesis explores some location and navigation method to solve these problems using computer vision technology.Specific research content and results include:1.A motion deviation detection algorithm based on vanishing point is proposed to solve the problem of the direction deviation of the unmanned car.In this thesis,the imaging characteristics of indoor passageway environment are studied.We studied the relation between deviation of unmanned car and vanishing point of parallel lines in passageway.Taking advantage of these imaging features,we proposed an algorithm to detect the motion deviation of the unmanned vehicle.In order to increase the speed of detection,we predict the range of line parameters of post frame using the result of preceding frame.We made experiments to prove that this algorithms performs well.The average processing time for a 640x480 image is less than 20ms.The accuracy is more than 98%.2.Location algorithm based on the feature of floor gap line.We studied the char-acteristics of floor gap line under the view of overlooking.By converting the images captured by the unmanned car camera to the overlooking angle,the floor gap lines can be detected quickly.The motion of the unmanned vehicle is cal-culated in real-time using the match of floor gap lines between images.Because of image blur and object occlusion,the matching chain could be broken.We proposed an algorithm to rejoin the matching chain by using the data of wheel odometer.3.Door detection algorithm based on line segment features.In this thesis,the line segments of door are studied.Base on previous research results,a line segment model of door is proposed.Doors are detected in real-time using the line seg-ments.The detection method is robust.Experiments show that it cost 44.73ms to process a image in size of 640×480.The accuracy rate is 89.4%.4.unmanned vehicle experiment platform.In order to test the algorithms,we set up an unmanned vehicle.It is used in test of location algorithm in outdoor and indoor environment.