Study Of Intelligent Robot On Simultaneous Localization

Simultaneous localization is an important part of simultaneous localization and mapping(SLAM).It is the key technology needed for intelligent robots to realize autonomous action,real-time positioning navigation and map construction,and it is also an important basis for the development of intelligent robot technology to a higher level.The basic principle of intelligent robots for simultaneous localization is that the robot determines the position and posture of oneself relative to the surrounding environment in real time by acquiring information about the surrounding environment.According to the simultaneous localization requirements of intelligent robots in complex indoor environment,this paper combines deep learning technology with the traditional visual SLAM to study the simultaneous localization of intelligent robots.The content of this paper is arranged as follows.1.The first part studies the acquisition methods and processing methods of visual information.Firstly,various visual sensors are introduced,and their advantages and disadvantages are compared.Depth camera is selected as the acquisition equipment of visual information according to the comparison results.Then,the depth camera is modeled to construct a mapping relationship between the object and the image.Several image features are introduced in details,and the ORB(Oriented FAST and Rotated BRIEF)features with low computational complexity and robustness are selected by comparison.Finally,the experimental feature are extracted and matched using the ORB feature.2.The second part studies the motion model of the camera carried by the robot and the estimation method of the camera pose.Firstly,the motion process of the intelligent robot is analyzed and the corresponding camera motion model is established.Then,the position and posture estimation of the camera is studied according to the motion model.The specific method is using image and its corresponding depth information,which captured by the depth camera under different camera position and posture,coordinate information of image feature points extracted under different camera position and posture is obtained,and to estimate the position and posture change of the camera according to the geometric relationship of the matched feature points.Finally,two camera position and posture estimation methods,linear estimation and nonlinear estimation,are introduced.The linear estimation with low computational complexity is selected as the position and posture estimation method through experimental comparison.3.The third part aiming at the problem that traditional visual SLAM will cause significant camera position and posture estimation error due to the interference of feature points belonging to the moving target,an optimization method of position and posture estimation based on pedestrian detection is proposed to reduce the estimation error.Firstly,a pedestrian detection networks model based on convolutional neural networks are designed and trained.The pedestrian detection and location algorithm is designed by using the trained networks model.Then,the method of camera position and posture estimation is improved.The specific operation is to remove the feature points belonging to the pedestrian image area,match the features and estimate the camera pose.And the above algorithm is validated by experimental images to prove the feasibility and validity of the method.Finally,according to the specific motion of intelligent robots and the requirements of low computational complexity and high estimation accuracy,the SLAM front-end is designed by using the theory and algorithm in this paper.And the front-end of the design is simulated.