Video stabilization refers to the problem of transforming a shaky video into a visually pleasing one. The question of how to strike a good trade-off between visual quality and computational speed has remained one of t...
Video stabilization refers to the problem of transforming a shaky video into a visually pleasing one. The question of how to strike a good trade-off between visual quality and computational speed has remained one of the open challenges in video stabilization. Inspired by the analogy between wobbly frames and jigsaw puzzles, we propose an iterative optimization-based learning approach using synthetic datasets for video stabilization, which consists of two interacting submodules: motion trajectory smoothing and full-frame outpainting. First, we develop a two-level (coarse-to-fine) stabilizing algorithm based on the probabilistic flow field. The confidence map associated with the estimated optical flow is exploited to guide the search for shared regions through backpropagation. Second, we take a divide-and-conquer approach and propose a novel multi-frame fusion strategy to render full-frame stabilized views. An important new insight brought about by our iterative optimization approach is that the target video can be interpreted as the fixed point of nonlinear mapping for video stabilization. We formulate video stabilization as a problem of minimizing the amount of jerkiness in motion trajectories, which guarantees convergence with the help of fixed-point theory. Extensive experimental results are reported to demonstrate the superiority of the proposed approach in terms of computational speed and visual quality. The code will be available on GitHub.
作者:
Guang-Song HanZhi-Hong GuanJie ChenDing-Xin HeMing ChiCollege of Automation
Huazhong University of Science and Technology Wuhan 430074 China and the Key Laboratory of Image Information Processing and Intelligent Control (Huazhong University of Science and Technology) Ministry of Education Wuhan 430074 China
A multi-tracking problem of multi-agent networks is investigated in this paper where multi-tracking refers to that the states of multiple agents in each subnetwork asymptotically converge to the same desired trajector...
详细信息
A multi-tracking problem of multi-agent networks is investigated in this paper where multi-tracking refers to that the states of multiple agents in each subnetwork asymptotically converge to the same desired trajectory in the presence of information exchanges among *** multi-tracking of first order multi-agent networks with directed topologies was ***-triggered protocols were proposed along with triggering functions to solve the stationary multi-tracking and bounded dynamic *** self-triggered scheduling is obtained, and the system does not exhibit Zeno *** examples are provided to illustrate the effectiveness of the obtained criteria.
In this paper, we propose an Activity-List based Nested Partitions algorithm for solving the Resource-Constrained Project Scheduling Problem(RCPSP). This algorithm is based on traditional Serial Scheduling scheme (SSS...
详细信息
In this paper, we propose an Activity-List based Nested Partitions algorithm for solving the Resource-Constrained Project Scheduling Problem(RCPSP). This algorithm is based on traditional Serial Scheduling scheme (SSS) and partitions the feasible solution space which is formulated by activity-lists into subregions by the nested partitions approach. We also utilize Double Justification as local search to improve the solutions. The algorithm is tested on J120 in PSPLIB with the result that the algorithm is relatively effective for solving large-scale, complex RCPSPs.
The analytical algorithm of program quaternion is studied, aiming at the problem of the arbitrary space vehicle attitude-adjusting control. The analytical constructor method of the program quaternion is provided for a...
详细信息
In this letter, a class of complex dynamical networks with additive stochastic time-varying delays is investigated. Two kinds of delays in complex dynamical networks are taken into consideration, one is called the nod...
详细信息
Due to the limitation of the vertical angle resolution of the lidar sensor,the observation information in the vertical direction is reduced,resulting in the elevation cumulative drift phenomenon of many lidar SLAM alg...
详细信息
ISBN:
(数字)9789887581581
ISBN:
(纸本)9798350366907
Due to the limitation of the vertical angle resolution of the lidar sensor,the observation information in the vertical direction is reduced,resulting in the elevation cumulative drift phenomenon of many lidar SLAM algorithms in large-scale environment,which reduces the accuracy of pose estimation and the quality of *** order to solve this problem,based on LIO-SAM,this paper proposes a lidar SLAM algorithm with ground plane *** method can effectively extract the ground point cloud and obtain the plane normal vector,and construct the constraint relationship with the prior plane normal *** optimal state estimation is solved by nonlinear optimization method,which effectively reduces the elevation drift caused by the long-term operation of the system in large-scale *** paper compares the improved algorithm with A-LOAM,LeGO-LOAM and original LIO-SAM algorithms in challenging datasets such as KITTI,M2 DGR and *** results show that this method effectively improves the accuracy of pose estimation and the quality of mapping,and limit the elevation drift of mobile robot in ground operation,which has high application value.
For battery management systems, it is significant to reliably estimate state-of-charge (SOC) from limited measurements in real time. Based on a nonlinear SOC-dependent equivalent circuit model, we propose a real-time ...
详细信息
ISBN:
(纸本)9781665401166
For battery management systems, it is significant to reliably estimate state-of-charge (SOC) from limited measurements in real time. Based on a nonlinear SOC-dependent equivalent circuit model, we propose a real-time moving horizon estimation (MHE) algorithm for SOC estimation. For efficient computation, the sequential quadratic programming strategy is applied, and each quadratic programming (QP) iteration is solved via the structure-exploiting forward and backward recursions. To account for the limited computation time allowed within each sampling interval, the proposed algorithm does not fully solve the receding horizon optimization problem over each sliding window, but performs fixed number of QP iterations. The simulation results show that compared to the existing fully solved MHE applied to SOC estimation, the proposed real-time MHE algorithm significantly saves computation time at the cost of a slight loss of estimation performance.
The characterization and understanding of online social network behavior is of importance from both the points of view of fundamental research and realistic utilization. In this manuscript, we propose a stochastic dif...
详细信息
3D object detection plays a critical role in autonomous driving perception. Although multi-view-based perception solutions have made significant progress, their performance is still far from being ready for practical ...
详细信息
3D object detection plays a critical role in autonomous driving perception. Although multi-view-based perception solutions have made significant progress, their performance is still far from being ready for practical use. The estimation of pixel depth is dependent on camera intrinsic properties, which led us to explore a depth-aware model guided by camera parameters. Our contribution in this paper is the DAFormer, which incorporates camera parameters and position-aware image features to detect 3D objects. The depthaware module uses camera parameters to reweight image features and estimate depth, enhancing and offsetting the 3D position embedding. The object query uses depth information enhanced features for end-to-end 3D detection by Attention layers. DAFormer achieves impressive results on the standard nuScenes dataset without any additional embellishments.
Active learning (AL) selects the most beneficial unlabeled samples to label, and hence a better machine learning model can be trained from the same number of labeled samples. Most existing active learning for regressi...
详细信息
暂无评论