Research On Mapping And Navigation Method Based On Binocular Vision

With the ever-changing production environment and product demand,as well as the rising labor costs,the company's production model shifts toward a flexible and intelligent production model.The logistics transportation system consisting of Automated guided vehicles(AGV)is an important part of intelligent manufacturing,and has a profound impact on logistics efficiency and costs.The distributed multi-agent organizational structure provides a solution to the challenges brought by the increasing production scale.However,it also puts higher requirements on the intelligence level of the vehicles.This paper takes a single vehicle as the research object and conducts research on the key technologies in its navigation method.The work done is as follows.Firstly,an overall analysis of the automatic guided vehicle system was carried out,and the overall framework of the AGV navigation system was described.Then the relevant background knowledge and technology to be used are deduced,analyzed or introduced,including camera model and distortion,the basic principle and implementation of visual SLAM based on graph optimization,the representation method of environment map and the related knowledge of parallel computing programming.A map that reflects the external environment is the basis for navigation.This paper studies the mapping method based on binocular stereo vision.The basic principle of binocular stereo matching is analyzed in detail.A stereo matching algorithm is designed and implemented on the GPU device.The accuracy and speed of the algorithm are tested on the data set.Finally,the algorithm is integrated into the original framework of ORBSLAM2 by adding a dense mapping thread to insert the generated point cloud at the key frame pose to create a continuous and dense map.Experiments show that the algorithm and system designed in this paper can create a map in real time,and the created map can effectively reflect the real environment of the outside world.This paper studies the application of genetic algorithm in path planning.Genetic algorithms are subject to complex calculations and are not suitable for real-time applications.The article analyzes the basic principles of genetic algorithms and summarizes the parallelism of genetic algorithms.Aiming at the above deficiencies,the implementation details of coarse-grained parallel genetic algorithm for path planning are studied in detail,and the performance of parallel models with different granularities under different computing platforms is explored.Finally,experiments show that the genetic algorithm designed in this paper that uses the coarse-grained parallel model calculating on multi-core CPU,which can make full use of computing resources,speed up the algorithm operation and improve the premature convergence.Applying this algorithm to path planning can achieve good planning effect and acceleration.