Research On Pose Estimation Method Based On Multi-vision Sensor Information Fusion

The pose estimation is a method of acquiring the pose and posture of the target in space by acquiring the feature information using the sensor.The accurate acquisition of the pose plays a key role in the application of accurate tracking of the manipulator,attitude adjustment of the spacecraft,determination of the position of the surgical instrument,attitude measurement of the wind tunnel model and so on.Vision sensor plays an important role in pose estimation system because of its low cost and high accuracy of pose estimation.However,there are a lot of problems in practical applications,such as glare,external noise interference,human intervention and so on,which often results in the incomplete feature information acquired by the sensor,and then influences the estimation result of the system.This paper focuses on the technologies such as vision-based pose estimation,multi-sensor,information fusion and occlusion.The main contents are as follows:1.An unscented Kalman filter pose estimation method based on adaptive adjustment of process noise variance is proposed.For the problem that the system process noise is unknown in the pose estimation process,this paper proposes a method for adaptive adjustment of process noise variance for unscented Kalman pose estimation.By introducing a modified Sage-Husa noise estimator,the process noise covariance is adjusted during each filter iteration,which reduces the impact of the process noise variance uncertainty on the estimation results.The simulation and experimental results show that the adaptive unscented Kalman filter pose estimation method can effectively estimate the target pose and improve the pose estimation accuracy.2.A matrix weighted distributed fusion pose estimation method based on adaptive unscented Kalman filter is proposed.Aiming at the problem of high noise sensitivity and insufficient estimation accuracy of the single vision sensor pose estimation system,this paper constructs a multi-sensor pose estimation system by using multiple vision sensors.First,each subsystem is estimated by the proposed adaptive unscented Kalman filter method,and then the estimation result of each subsystem is fused by distributed fusion method,so that the estimation accuracy of the system is higher than any result of the subsystems.The simulation and experimental results show that the matrix weighted distributed fusion pose estimation method proposed in this paper improves the estimation accuracy of the system to some extent,and the pose estimation result is better than any subsystem.3.An adaptive unscented Kalman distributed fusion method for occlusion is proposed.Aiming at the problem that the visual sensor is easily affected by the external environment which result in incomplete information acquired by the sensor,this paper proposes a method based on adaptive unscented Kalman distributed fusion to deal with the occlusion.The subsystems are divided into three categories based on the amount of characteristic information that can be observed: severe occlusion subsystem,partial occlusion subsystem and normal subsystem.And corresponding processing methods are proposed for each category.The experimental results show that the processing method proposed in this paper can ensure the stable operation of the system and ensure the accuracy of pose estimation when the occlusion subsystem appears in the multi-visual sensor pose estimation system.