Dissertation
Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device

Replay-Based Soccer Video Highlights Detection

Author ZhangQiang
Tutor GaoGuangRong;YuJunQing
School Huazhong University of Science and Technology
Course Applied Computer Technology
Keywords Event Detection Playback scenes Emotional motivation model Goal detection
CLC TP391.41
Type Master's thesis
Year 2011
Downloads 12
Quotes 0
Download Dissertation

Due to the presence of the the Internet Shanghai amount of soccer video , fast-paced life makes most concerned by the user , not how to transfer and play all the games , but out of concern their own interesting or exciting video clips . Therefore, how to analyze a football video content to meet the needs of different equipment , different needs of the audience has a very important significance . Extraction and semantic event detection method based on systematic analysis of soccer video shots , playback scene and goal detection , an effective solution to the existing emotional incentive model can not respond well to the intensity of the game , the wonderful lens positioning and event detection enough ideal problem. Shot boundary detection , improved shear and gradient lenses discriminant proposed shot boundary detection algorithm based on statistical methods to confirm the gradual football . Playback scene detection based Logo Logo access : for convert video movement characteristics Logo , the similarity by motion vector the contrast automatically obtain Logo pixels set ; conversion complex video , design for Logo only need a small amount of interaction can be extracted Logo pixel set handy tool . On the detection of the goal , there is constructed a feature vector containing the goalpost width , and using SVM recognition goal image . Through the playback scene features improved emotional incentive model , and the short, medium and long three different shots boundary location . To achieve a goal , shot and foul event rules wonderful lens semantic event detection and the emotional excitation curve and event detection applied to search net . Experimental results show that the proposed playback scene detection methods have very high recall rate and accuracy . The playback scene features emotional excitation curve more accurately reflects the fierce football game , semantic event detection , especially scoring event , has a higher recall rate and accuracy . However , for some special playback effects require further processing , the rules of the shot and foul events need to be further strengthened in order to improve the recognition accuracy .

Related Dissertations
More Dissertations