Research And Implementation Of Obstacle Recognition Technology Based On Multi-sensor Data Fusion

In recent years,with the continuous development of intelligent technology,people put forward higher requirements for the autonomy of robots.It is expected that they can not only complete actions according to human instructions,but also have certain autonomy in complex environments.In an unknown environment,robots want to have autonomy,the first task is to accurately perceive the surrounding environment,so using sensor information to detect obstacles is one of the important research directions in the field of robots.However,each sensor has its own limitations,such as laser sensor is insensitive to transparent objects,visual sensor can not work in dark environment,so it is difficult to achieve the accuracy and stability requirements of the detection system by relying on a single sensor.In order to solve this problem,this paper proposes an obstacle recognition technology based on multi-dimensional heterogeneous sensor data fusion.By fusing the complementary information of multiple sensors,obstacles are detected from different angles and aspects,and an appropriate tracking model is established to avoid the failure of a single sensor in a specific environment,thus improving the obstacles.The reliability and accuracy of the system.In order to solve the problem of detection failure caused by inherent limitations of sensors,this paper proposes to break through the inherent limitations of a single sensor by using multi-sensor and multi-angle complementary information,so as to improve the accuracy and stability of obstacle detection.Around this research goal,this paper proposes a barrier recognition technology based on multi-sensor data fusion,including two main research contents: obstacle recognition framework supporting multi-sensor automatic calibration mechanism and obstacle tracking mechanism based on data fusion and association technology.Finally,an experimental simulation is carried out in Gazebo simulator.It's true.The effectiveness of the algorithm is verified by many experiments.The main work of this paper includes the following aspects:(1)To solve the problem of multi-sensor data alignment,this paper proposes a calibration mechanism with strong generalization ability,which aligns visual data and lidar data in time and space.Its characteristic is that it does not need complicated pre-setting,and can complete the calibration process independently without human intervention.The experimental results show that the algorithm is independent of sensor parameter setting and surrounding environment,and has high generalization.This mechanism unifies the visual data and the lidar data,and provides guarantee for the obstacle detection part.(2)In order to improve the accuracy of obstacle detection,a two-tier distributed Kalman filter is proposed to fuse the data of multiple sensors.Compared with the centralized Kalman filter,it is more fault-tolerant.Then it improves the current mainstream data association algorithm.Compared with the traditional PDA algorithmand JPDA algorithm,the improved algorithm has greater advantages in time and accuracy.Finally,based on the above estimation and association algorithm,this paper proposes a new strategy for track deletion and creation,defines two types of tracks,and explains their deletion and update conditions.(3)In order to verify the effectiveness of the tracking algorithm,based on the above technology,this paper implements the system prototype in Gazebo simulator and carries out simulation experiments.After repeated experiments,the results show that the proposed recognition framework based on multi-sensor data fusion technology has high generalization calibration ability and supports multiple sensors.Obstacle detection can effectively reduce missing detection rate and false detection rate.