Zhimin Fan

Multiple collaborative kernel tracking

Those motion parameters that cannot be recovered from image measurements are unobservable in the visual dynamic system. This paper studies this important issue of singularity in the context of kernel-based tracking and presents a novel approach that is based on a motion field representation which employs redundant but sparsely correlated local motion parameters instead of compact but uncorrelated global ones. This approach makes it easy to design fully observable kernel-based motion estimators. This paper shows that these high-dimensional motion fields can be estimated efficiently by the collaboration among a set of simpler local kernel-based motion estimators, which makes the new approach very practical.

Multiple Collaborative Kernel Tracking

Those motion parameters that cannot be recovered from image measurements are unobservable in the visual dynamic system. This paper studies this important issue of singularity in the context of kernel-based tracking and presents a novel approach that is based on a motion field representation which employs redundant but sparsely correlated local motion parameters instead of compact but uncorrelated global ones. This approach makes it easy to design fully observable kernel-based motion estimators. This paper shows that these high-dimensional motion fields can be estimated efficiently by the collaboration among a set of simpler local kernel-based motion estimators, which makes the new approach very practical.

Efficient Optimal Kernel Placement for Reliable Visual Tracking

This paper describes a novel approach to optimal kernel placement in kernel-based tracking. If kernels are placed at arbitrary places, kernel-based methods are likely to be trapped in ill-conditioned locations, which prevents the reliable recovery of the motion parameters and jeopardizes the tracking performance. The theoretical analysis presented in this paper indicates that the optimal kernel placement can be evaluated based on a closed-form criterion, and achieved efficiently by a novel gradient-based algorithm. Based on that, new methods for temporal-stable multiple kernel placement and scale-invariant kernel placement are proposed. These new theoretical results and new algorithms greatly advance the study of kernel-based tracking in both theory and practice. Extensive real-time experimental results demonstrate the improved tracking reliability.

Tracking Nonstationary Visual Appearances by Data-Driven Adaptation

Without any prior about the target, the appearance is usually the only cue available in visual tracking. However, in general, the appearances are often nonstationary which may ruin the predefined visual measurements and often lead to tracking failure in practice. Thus, a natural solution is to adapt the observation model to the nonstationary appearances. However, this idea is threatened by the risk of adaptation drift that originates in its ill-posed nature, unless good data-driven constraints are imposed. Different from most existing adaptation schemes, we enforce three novel constraints for the optimal adaptation: (1) negative data, (2) bottom-up pair-wise data constraints, and (3) adaptation dynamics. Substantializing the general adaptation problem as a subspace adaptation problem, this paper presents a closed-form solution as well as a practical iterative algorithm for subspace tracking. Extensive experiments have demonstrated that the proposed approach can largely alleviate adaptation drift and achieve better tracking results for a large variety of nonstationary scenes.

Multibody grouping by inference of multiple subspaces from high-dimensional data using oriented-frames

Recently, subspace constraints have been widely exploited in many computer vision problems such as multibody grouping. Under linear projection models, feature points associated with multiple bodies reside in multiple subspaces. Most existing factorization-based algorithms can segment objects undergoing independent motions. However, intersections among the correlated motion subspaces will lead most previous factorization-based algorithms to erroneous segmentation. To overcome this limitation, in this paper, we formulate the problem of multibody grouping as inference of multiple subspaces from a high-dimensional data space. A novel and robust algorithm is proposed to capture the configuration of the multiple subspace structure and to find the segmentation of objects by clustering the feature points into these inferred subspaces, no matter whether they are independent or correlated. In the proposed method, an Oriented-Frame (OF), which is a multidimensional coordinate frame, is associated with each data point indicating the points preferred subspace configuration. Based on the similarity between the subspaces, novel mechanisms of subspace evolution and voting are developed. By filtering the outliers due to their structural incompatibility, the subspace configurations will emerge. Compared with most existing factorization-based algorithms that cannot correctly segment correlated motions, such as motions of articulated objects, the proposed method has a robust performance in both independent and correlated motion segmentation. A number of controlled and real experiments show the effectiveness of the proposed method. However, the current approach does not deal with transparent motions and motion subspaces of different dimensions.

Multibody motion segmentation based on simulated annealing

The problem of multibody motion segmentation is an important and challenging issue in computer vision. In this paper, a novel segmentation technique based on simulated annealing (SA) is proposed. According to the fact that under linear projection models, feature points of multibody reside in multiple subspaces, firstly, a meaningful energy function is proposed, which favors the correct formation of those subspaces, and some subspaces are generated as the initial state. Then, two strategies of subspace evolution and transformation are developed to optimize the energy function in a manner of simulated annealing. The ultimate configuration of these subspaces will reveal the inherent multiple subspace structure embedded in the data space. The classification of data points to these subspaces is equivalent to multibody grouping. The global optimization process results in an increase of robustness with noise tolerance. The method is also effective in degenerate cases. Promising results on synthetic and real data are presented.

Robust contour extraction for moving vehicle tracking

A robust framework is proposed for contour extraction and moving vehicle tracking. First, we establish a modified snake model and utilize directional information to guide the behavior of snaxels. Then, an adaptive shape restriction is embedded into the algorithm to govern the scope of the snakes motion. The spatio-temporal relationship between successive frames is estimated using a Kalman filter. These can improve the snakes robustness against noise or the presence of occlusion, which is inevitable in real tasks of traffic monitoring. Experimental results of the proposed framework on real traffic videos are satisfying and encouraging.

Inference of multiple subspaces from high-dimensional data and application to multibody grouping

Multibody grouping is a representative of applying sub-space constraints in computer vision tasks. Under linear projection models, feature points of multibody reside in multiple subspaces. We formulate the problem of multi-body grouping as multiple subspace inference from high- dimensional data space. The theoretical value and practical advantage of this formulation come from the relaxation of the motion independency assumption, which has to be enforced in most factorization, based methods. In the proposed method, an oriented-frame (OF), which is a multi-dimensional coordinate frame, is associated with each data point indicating the points preferred subspace structure. Then, a similarity measurement of these OFs is introduced and a novel mechanism is devised for conveying the information of the inherent subspace structure among the data points. In contrast to the existing factorization-based algorithms that cannot find correct segmentation of correlated motions such as articulated motion, the proposed method can robustly handle motion segmentation of both independent and correlated cases. Results on controlled and real experiments show the effectiveness of the proposed sub-space inference method.

A new model to forecast the results of matches based on hybrid neural networks in the soccer rating system

The objective of this paper is to build a result prediction model for the rating system in soccer games. A rating system which plays a crucial role in world sports field yields predictions for the probability that one contestant beats another. The result prediction model is the core technique in the rating system. The robustness and accuracy of the model is a very important feature because people will trust the rating system only if it can give the exact prediction of the game results. This paper employs a coarse-to-fine training technique based on hybrid neural network. Very few people have ever attempted the method based on neural network before in this field. First a match is classified into three categories with a LVQ net to determine the strength contrast between two contestants. Then the elaborately designed data will go through the specific BP nets according to the classifying result. The model is trained and tested on volumes of actual soccer match results from Italian series A. Finally the results of the model are compared to other prediction models based on statistics. The outcome shows that the new model is more accurate and provides better performance evaluation of all teams.