Optimizing precision of SIFT algorithm in feature extraction of tennis video
Changhong Wu1, Haiyan Geng1, Tianlin Geng2
COMPUTER MODELLING & NEW TECHNOLOGIES 2014 18(12B) 559-564
1College of Sports, LangfangTeachers College, Langfang, 065000, Hebei, China
2BaodingGaoyangHongrunMiddle School,Baoding, 071500, Hebei, China
The traditional SIFT algorithm still has problems such as running slowly and low accuracy in the tennis video feature extraction and matching, an improved SIFT algorithm is proposed based on a tennis video feature extraction and matching. First, it limits the number of feature points to SIFT algorithm by adding the image texture features, which make the feature points to evenly distribute in each set of different scales of video image. Then measures the similarity of feature points by Euclidean distance, the measurement results transform with projection transformation relations, and then uses iterative arithmetic of random sampling consistency (RANSAC) algorithm to obtain maximum satisfy feature points of geometry model. Finally, uses the minimum root meansquare error (RMSE) to determine the accuracy of registration. The simulation experiments show that the proposed improved SIFT algorithm based on tennis video feature extraction and matching has faster running speed and better matching precision.