

Articles
by
Yaming Wang 
Total Records (
7 ) for
Yaming Wang 





Yaming Wang
and
Zhou Xu


3D elastic motion estimation from monocular image sequence is crucial for many important applications. In this paper, an approach is proposed for 3D elastic motion from monocular image sequence using vectorentropy regularization. First, with the established correspondences of feature points between consecutive frames, the least squares estimation model is proposed based on affine motion model and central projection model. Then, in order to overcome the illposed 3D motion estimation problem, a method using regularization is proposed. A vector entropy consisting of the second order entropy (Ent2) and the cross entropy is constructed as the regularization term which incorporates the prior motion knowledge into the estimation process. By imposing the motion constraints, the vectorentropy regularization converts the illposed problem into a wellposed one and guarantees the robust solution. Experimental results from a synthetic image sequence demonstrate the feasibility of the proposed approach.





Yaming Wang
and
Yuanmei Wang


A neural network based approach to feature point correspondence in a long image sequence is proposed in this paper. The proposed approach formulates the correspondence problem as a constrained optimization problem and uses a Hopfield neural network to find the solution. The problem is viewed as that of optimizing a suitable energy function which is chosen by taking into account attributes including Euclidean distance between two feature points in two consecutive frames, prediction of feature points using filter, occlusion of points, and smoothness of point motion. This approach is robust especially for occlusion. At the same time, this method provides the twodimensional motion parameters, so it may be used in twodimensional motion analysis and predictive visual tracking. Experimental results on a synthetic sequence and a real image sequence of human motion show that the correspondence can be established efficiently, although it has been previously proved to be a combinatorially explosive problem. 




Yaming Wang
,
Yunhua Zhang
,
Li Cao
,
Weida Zhou
and
Wenqing Huang


To segment the moving objects from image sequence is an important problem in computer vision. In this paper, a novel approach is proposed for moving objects segmentation from grayscale image sequence. In order to increase the robustness of segmentation, the informationtheoretic principle of mutual information is proposed to combine two different segmentation results. First, the adaptive Gaussian distribution model for each image pixel is presented. Based on this model, each image frame is mapped from spatial domain to statistic domain. The segmentation is consequently performed in statistic domain. Meanwhile the image frame is segmented using Otsu`s method (Otsu, N., 1979) This two segmentation results are then combined by using mutual information and the final result is obtained. Experimental results from a real image sequence show the feasibility of the proposed approach. 




Yaming Wang
,
Weida Zhou
and
Xiongjie Wang


This study proposes a novel approach based on Cellular Neural Networks (CNN) is proposed for segmenting moving objects from monocular image sequence regardless of complex, changing background. First, a Gaussian distribution model for image pixel is proposed. The parameters contained in the model are adaptively updated based on the information from the current and historical frames. Based on this, every image frame is mapped from spatial domain to statistical domain. Then, a CNN framework is proposed for segmenting moving objects in statistical domain. The desirable feature of CNNs is that the processors arranged in the two dimensional grid only have local connections, which lend themselves easily to VLSI implementations. By modeling pixel interactions through using a spatialtemporal neighborhood of the CNN, sparse nosy pixel can be erased and robust segmenting results of moving objects can be achieved. Experimental results from two real monocular image sequences demonstrate the feasibility of the proposed approach. 




Yunhua Zhang
,
Yaming Wang
,
Sun Li
and
Wenqing Huang


This study uses neural networks to estimate threeDimensional (3D) rigid motion parameters based on twoDimensional (2D) motion fields. The motion fields are computed from image sequences. The neural networks update their weights by NewtonRaphson procedure for minimizing the error measures. Experimental results are presented for validating the proposed approach. 





Yunhua Zhang
,
Yaming Wang
and
Wenqing Huang


A novel approach to 3D motion estimation of elastic body from monocular image sequence is
proposed in this paper. First, with the establishment of feature point correspondence between consecutive
image frames, the affine motion model and the central projection model are presented for local elastic motion.
Then, in order to obtain the global motion parameters and overcome the illposed 3D estimation problem, a
framework of Markov Random Field (MRF) with entropic constraints is proposed. By incorporating the motion
prior constraints into the MRF, the motion smoothness feature between local regions is reflected. This converts
the illposed problem into a wellposed one and guarantees the robust solution. Experimental results from a
sequence of synthetic image sequence demonstrate the feasibility of the proposed approach.





Yaming Wang
,
Li Sun
,
Jueliang Hu
,
Yunhua Zhang
and
Wenqing Huang


3D rigid motion estimation from images is crucial for many important applications. In this study, an approach is proposed for 3D rigid motion estimation from feature points using feedforward neuralnetworks. The correspondence of feature points between consecutive images is assumed to be established beforehand. The proposed neural network is composed of 3 layers and 3 points are randomly selected from all points on the object to train the network using NewtonRaphson procedure. Experimental results from synthetic data are presented for validating the proposed approach. 





