Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2022-06-06 Cooperative journals: 《计算机应用研究》
Abstract： Pose graph optimization (PGO) is a high dimensional non-convex optimization algorithm widely used in the field of computer vision. It is hard to solve directly and mainly relies on iterative techniques. It requires high quality of initial value which is difficult to be guaranteed in practice. This paper studied PGO problem and proposed a simple closed solution algorithm for pose graph based on eigen decomposition. This algorithm first developed the semidefinite relaxation of maximum-likelihood estimation (MLE) for PGO problems. Then it transformed MLE into eigen decomposition problem, and designed an improved model reduction method to solve the problem by using the sparsity of data, which further improves the computational speed of the algorithm. The algorithm has the advantages of scalability, low computational cost and high precision. Finally, experimental evaluation on simulated and real-world pose-graph datasets shows that the proposed algorithm can optimize the pose graph quickly without compromising accuracy.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2022-05-10 Cooperative journals: 《计算机应用研究》
Abstract： In order to improve the accuracy and efficiency of the feature point matching, this paper proposed a novel feature point matching algorithm in terms of vision and inertial measurement unit (IMU) fusion. Firstly, the algorithm calculated the point diffusion function using the motion information of IMU to remove motion blur, and improved the feature point matching rate. Secondly, based on LK (Lucas-Kanade) optical flow method, this paper introduced gradient error and uses L1 parametric to simulate sparse noise. Furthermore, the feature point position by using IMU is the initial value of the algorithm, and then this paper used BB (Barzilar-Borwein) step to improve the efficiency of the algorithm. Finally, the comparison experiments show that the efficiency and accuracy of the algorithm are better than the LK optical flow method. Especially, the algorithm improves the localization accuracy and robustness of the VINS-Mono framework on the dataset EuRoC.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2019-04-01 Cooperative journals: 《计算机应用研究》
Abstract： This paper proposed an uncalibrated 3D robotic arm grabbing method based on improved YOLOv2 in a multi-object environment. Firstly, in order to reduce the depth learning algorithm YOLOv2 detection multi-object bounding box overlapping rate and 3D distance calculation error. It proposed an improved algorithm for YOLOv2. Using this algorithm to detect and identify the target object in the image, obtain the position information of the target object in the RGB image, and then use the k-means++ clustering algorithm to quickly calculate the distance from the target object to the camera according to the depth image information, and estimate the target object size and pose. Simultaneously, use the improved YOLOv2 to get the bounding box of the gripper and calculate the distance from the robot to the target object. Then the system estimates the distance between the fixture, camera and object in the manipulator coordinate system. Finally, the system uses the PID algorithm to control the gripper to grab the object according to the size and posture of the object and the distance from the object to the gripper. In this paper, the detected boundary boxes of the target object is more accurate based on the improved YOLOv2 than on old one. It also enhances the distance from the fixture to the object and the size of the object as well as the accuracy of the pose estimation. In addition, in order to avoid complicated calibration, this paper proposes a non-calibration method. This learning scheme is different from the traditional uncalibrated estimation method based on Jacobian matrix, because it has good universality. A simulation experiment shows that the proposed method can accurately classify and locate the objects in the image, The Universal Robot 3 robotic arm uses this framework to verify the effectiveness of capturing objects in a cluttered environment.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-19 Cooperative journals: 《计算机应用研究》
Abstract： Constraint satisfaction problem is an important issue in the field of artificial intelligence. This paper proposed two algorithms combining belief propagation and simulated annealing to solve a random constraint satisfaction problem with exact phase transitions and large number of hard instances. The algorithms firstly obtain the marginal probability distribution of variable values after the convergence of the belief propagation equation, then uses the strategy of maximum probability and minimum component entropy to generate a set of heuristic initial assignments, and then uses simulated annealing to modify the assignments. The experimental results show that the algorithms greatly improves the convergence rate from the initial assignments toward the optimal solution, and shows a significant advantage over simulated annealing algorithm.
Hits
Hits
Downloads
Comment