The Research Of Feature Point Detection And Matching Based On Three-dimensional Reconstruction

With the development of computer technology and the needs of practical, three-dimensional reconstruction techniques become a hot issue at home and abroad. Three-dimensional reconstruction use the primitives of two-dimensional image such as points, lines, surfaces, etc. to restore three-dimensional scene. Feature point detection and matching technology is the basic and most important work of three-dimensional reconstruction .Because the feature point detection and matching results will directly affect the follow-up.Harris corner detection algorithm, which is high precision, stability, and calculat- eon quickly, is based on the Moravec algorithm. It is more classical algorithm of corner detection algorithm . But the algorithm depends on the setting of the threshold in single scale detection feature points detection, and the threshold value is dependent on the properties of the image.Especially it is difficult to determine the threshold value because of color tones, so that it is bind for users to set the threshold value. you can only set the threshold value several times to get the relative ideal corner.then if the threshold setting unreasonable this will produce a corner point clustered and a corner point skewed, which will impact on the follow-up of Matching images. adaptive Corner Detection Algorithm appears for the shortcoming of Harris corner detection algorithm . this algorithm base on the original Harris corner detection algorithm, and block of the image using the thi- nking of fixed block, thus avoiding of set the threshold value by users and ensuring uniformity of point distribution ;Secondly ,we must remove the pseudo-feature point of image with SUSAN , so to prevent the corner clustering phenomenon.Scale invariant feature transform (SIFT) algorithm based on Local invariant des- criptor method is the more classical algorithm in all of the Image matching algorithm.But the algorithm should assign a main direction for the key point,that in order to ensure that the descriptor rotation invariance,and the operator in 128-dimensional feature point generates a greater impact on the efficiency of the algorithm, operator the improved SIFT algorithm generate a description operator in itself anti-rotation ability, and put the original 128-dimensional description operators reduced to 24 dimension ,which reduces the complexity of the algorithm.