The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
Because of the above factor it inescapable to cause the error 
matching. Although it takes lot of matching method to ensure 
the correspondence of moving object feature points such as 
taking the image correlation based feature, relaxation matching 
method based matching sustain, image relation based gray, and 
matching in local area etc, but there are still some error 
matching points. For example, in the above experiment, from 
the step CD to step®, the correct feature points ’ amount which 
obtained is decrease as the process. For example the feature 
points in CD feature extraction from same sequence image, there 
are 17 feature points, the right matching points are 15, that is the 
number, in the process®, of different sequence stereo matching 
feature points arel5, but after finishing the 3D object-side 
feature point correspondence, the correct number of feature 
points only are 13, the final correspondence is about 76.5%. So 
it add feature matching robust arithmetic in the program, such 
as using the mature bundle adjustment theory in 
photogrammetry field, to auto discover the error in feature 
matching or correspondence, and form a feed back control 
process, it worth to further study. 
4. CONCLUSIONS 
In order to realize the location and tacking of the three- 
dimensional object based on points feature, this paper discusses 
the most challengeable problem in this process that is the 
features correspondence problem. The algorithm of double 
restriction matching combining motion and stereo image 
matching is presented after analysis the character of sequence 
(motion) matching and stereo matching in the process. Take the 
advantages of high overlap of motion video frequency images, 
The final correspondence is about 75.6%, which is validated by 
real data experiment. It can meet the requirements of the 3D 
motion object tracking location. It will be the future study to 
give the further optimizing and robust algorithm of feature point 
correspondence. 
REFERENCES 
[ljR.Hartley.A.Zisserman Multiple View Geometry in 
Computer Vision [M] Cambridge University Press 2000.238- 
241 
[2] ZHANG Chun-sen. Research of 3-D Motion Object Tracking 
Location Based on Stereo Sequence Images [D], PhD 
Dissertation of Wuhan University, 2004(in Chinese) 
[3] Z.Zhang, R. Derche, O. Faugeras and Q. T. Luong, A robust 
technique for matching two uncalibrated images through the 
recovery of the unknown epipolar geometry[J].Artificial 
Intelligence 1995,78(1-2): 87-119. 
[4] MA Song-de,ZHANG Zheng-you. Computer Vision: 
Fundamentals of Computational Theory and Algorithms [M]. 
Beijing: Sciences Press, 1998:15 l-173(in Chinese) 
[5] Pui-Kuen Ho and R. Chung. Stereo-Motion with Stereo and 
Motion in Complement [J], IEEE Transactions on Pattern 
Analysis and Machine Intelligence, 2000,2(2):215-220 
[6] F. Domaika and R. Chung. Cooperative Stereo-Motion: 
Matching and Reconstruction [J], Computer Vision and Image 
Understanding, 2000, 79(3): 408-427 
[7] Zhengyou Zhang. Motion and Structure of Four Points from 
One Motion of a Stereo Rig with Unknown Extrinsic 
Parameters [Jj.IEEE Transactions on Pattern Analysis and 
Machine Intelligence, 1995, 17(12): 1222-1227 
Project supported by the National Science Foundation of China 
(No.40171080)