Research On Outdoor Multi-Sensor Fusion Algorithm Based On Visual SLAM

In recent years,Simultaneous Localization and Mapping(SLAM)technology based on vision has been widely recognized and applied in the fields of logistics distribution,autonomous flight and autopilot,SLAM system still faces many problems:sparse scene features make it difficult to extract image features,which can not guarantee the accuracy of pose estimation results;It is difficult to avoid the accumulative error in the longtime operation,and the real-time location results can not be obtained in the case of loop.Inertial Measurement Unit(IMU)and vision are complementary in nature and can modify the Inertial Measurement Unit estimation results of the vision.Global Navigation Satellite System(GNSS)is a Global sensor,it can provide non-drift result for visual positioning.Therefore,the multi-sensor fusion localization scheme based on vision,IMU and GNSS is proposed to realize the accurate location of multi-rotor UAV in outdoor environment.The main work of this paper is as follows:(1)A binocular SLAM system based on semi-direct method is proposed,which is improved on SVO algorithm.The binocular camera provides depth information and can update the depth points more quickly,it provides a good basis for map construction,using image processing schemes based on joint Histogram equalization and adaptive Gamma transform,in this paper,we propose a selective point-frame optimization processing strategy to obtain higher quality feature points,which makes the images with drastic changes in outdoor luminosity retain the image details without violating the gray-scale invariance hypothesis,better maintain map scene extensions.Experiments show that binocular SLAM system based on semi-direct method has better performance in real-time and accuracy.(2)A nonlinear optimization method based on the tight coupling of vision,IMU and GNSS is proposed,and the multi-sensor fusion localization scheme is constructed and initialized,and the visual residuals and IMU residuals are derived,the GNSS data is transformed into constraint information and fused into the global optimization function of the system,which makes full use of the correlation between sensors and uses the sensor data more effectively.In the nonlinear optimization scheme,the sliding window edge strategy is used to reduce the information of nonlinear optimization key frames and reduce the system computation.(3)the UAV platform based on PX4 flight control hardware and ROS software is built,and the multi-sensor fusion localization scheme is validated in data set and real scene,and compared with other algorithms.The experimental results show that the proposed algorithm has better accuracy and can adapt to the outdoor complex environment.