Research On Multi-level Semantic Map Construction Algorithm For Mobile Robots

With the in-depth study of artificial intelligence algorithm,it has become the mainstream research direction to apply it to the field of robot,so that the robot can solve various practical problems in the complex and changeable real environment.This paper takes mobile robot as the research object,focuses on the intelligent sensing technology for mobile robot,and realizes the detection and prediction of various scales of information in the surrounding environment by constructing multi-level semantic map.The main research work of this paper is as follows:1.Research on static semantic map construction algorithm.A rasterized semantic map construction algorithm based on semantic segmentation and lane detection was proposed to obtain the ground semantic information around the robot and update it in real time to construct the drivable area required by the path planning of the mobile robot.Moreover,the vectorization semantic map construction algorithm based on bird’s eye view features is studied to provide vectorization maps of lane lines and roadside lines for navigation tasks of mobile robots.2.Research on dynamic semantic map construction algorithm.The construction of dynamic semantic map is divided into 3D object detection and 3D object tracking.Based on the sensor configuration of monocular camera and laser,3D object detection algorithm based on visual-laser fusion and 3D object tracking algorithm based on location association and appearance association fusion are studied.Based on the sensor configuration of the circumnavigation multi-eye camera,the 3D object detection algorithm based on bird’s eye view features and the 3D object tracking algorithm based on deep learning speed prediction are studied,and the optimization of the algorithm in engineering are realized.And the management method of object tracking data is studied to realize the construction of dynamic semantic map.3.Research on trajectory prediction algorithm.The trajectory prediction algorithm is studied to predict the future trajectory of the target in the map and distinguish the driving intention,which provides information for the decision-making of the mobile robot.In this paper,the expression form of multilevel vectorized map and the conversion and sampling algorithm of vectorized instance are designed to realize the data form required for converting multilevel semantic map into trajectory prediction algorithm.In addition,a algorithm for finding interesting targets is designed to select important targets for prediction task.4.The related work of experiment.The mobile robot platform is designed for experiment.The simulation mobile robot platform and simulation of the world environment is designed for experiment.The off-line vectorization map generator in simulation environment is designed to meet the testing requirements of target trajectory prediction algorithm.A multi-level semantic map construction system is built on the physical platform and simulation platform and tested.