Research On Real-time Detection Technology Of Passable Area Based On Binocular Stereovision

Binocular stereovision technology uses binocular camera to simulate the visual function of human eyes,which achieves the construction,recognition and understanding of the objective world three-dimensional scene by indirectly obtaining two images of the same scene.The essence is to use the triangulation principle to calculate the true coordinate information of the left and right image points corresponding to the space object points.The main work of this thesis is based on the recognition characteristics of v-disparity map,and the research and system implementation of outdoor semi-structured pavement detection algorithm based on binocular vision are carried out.The algorithm of the detection system proposed in this thesis has more information and stronger plasticity.It can divide the area within the entire image range into a passable area and an impassable area.By using the advantages of binocular vision itself,the distance between the obstacle and the camera can also be theoretically detected.The overall system includes the design of the algorithm framework design and hardware platform construction.The following researches have been done based on the passable area detection technology:(1)Design and implementation of algorithmic framework for passable area detection system based on binocular stereovision Technology.The sparse matching based local feature algorithm is deeply studied and matching optimization is realized.It is compared with a global algorithm based on dense matching.The principle of construction of v-disparity map is studied in detail,and on this basis,a random sampling uniform line fitting algorithm and a road condition threshold classification algorithm are used to complete the analysis and feedback of the front traffic condition information.(2)Hardware platform and software development environment construction based on the embedded system.It elaborates the selection and connection principle of the hardware platform,including the binocular stereo vision system,embedded processing platform,feedback system and power supply system,as well as the construction method of the software compiling environment based on Linux operating system.The embedded platform used is an Odroid-xu4 processor from South Korea's Hardkernel company.In addition,this thesis also transplants the program to the Raspberry Pi 3 generation B embedded processor for testing and comparing the performance of the two processors.(3)Outdoor test experiments on the entire system.The application environment studied in this thesis is an outdoor environment,including accessible areas such as corridors,sidewalks,and ramps under strong and low light conditions.In this thesis,the calibration error analysis and distance error analysis of the binocular vision system are discussed in detail,and the detection results of the proposed algorithm frame in various scenes are presented,and the running speed of the system is tested on the computer and embedded platform respectively.Finally,all the data analysis and image effect are given.This thesis aims to build a set of embedded system which can be used for outdoor road detection,and gives a complete design scheme,including the design framework of the whole system,the way of hardware construction,the framework of algorithm flow,and so on,which not only establishes a complete theoretical system,but also uses the system used in real environment detection.Through the experiment,The proposed passable area detection technology to a certain extent solves the problem of poor adaptability of the environment.In the case of no sample training and learning,the purpose of real-time detection can be achieved on the premise of ensuring the recognition accuracy on the premise of the computer or the embedded platform.