Research Into Multi-object Identification Technology Based On Sequence Images

Multi-image sensor intersection measurement is a vital research part in computer vision. In multi-image sensor intersection measurement, it usually needed correlated information of the same object which was in different image sensors. Specially, when there were many objects that moving with random in space, due to the time and space location of every objects' correlated information changing in every time, at this time using different image sensors which measured by intersection measurement for any objects, it was very difficult.Aiming to this puzzle, the paper analyses and summarizes the object identification technology in image matching firstly, analyses the result of identification, and indicates the shortage of application in multi-image sensor.Later, to the drawback of multi-object identification in image matching, it researches the correlated identification technology which is between the location information of multi-image sensor with the correlated information of object image. By analyzing the camera imaging model in space, it uses the method of epipolar constrain to reducing wrong matching of corresponding points.On the base of the research above-mentioned, principally studies the estimating methods of epipolar geometry and fundamental matrix, based on which, a kind method that the whole least square estimation of RANSAC fundamental matrix estimation was proposed.Finally, the method applies in this paper is validated and analyzed. The result indicates that the improved fundamental matrix estimation can reduce wrong matching effectively, and can avoid the random of object identification. It realizes multi-object identification in image sequence well.