Analysis And Research Of Soccer Video Structuring

Video information and retrieval (VICR) is becoming an important issue in the field of multimedia with the increasing demand of customers to acquire needed information quickly by effective organization, representation, management, query and retrieval of the mass unstructurized video data, while traditional video information retrieval program fails to meet such a requirement in abstracting video content automatically, objectively and generally. Thus researches have begun to study content-based video retrieval (CBVR). However, CBVR is far from successful in bridging the semantic gap between low level features and high level semantics of video. That is to say, a lot of problems concerning semantic processing theory and technology of CBVR need a further study.The sports activity has attracted people all over the world with its own charm. Soccer is one of the most extensive sports activity, it is enjoyed by many fans. We find that every game of soccer video lasts a long time, but the things that really attract audience are little highlights shots. Most people tend to only enjoy their favorite games and fragments and they wish to find out the game fragments which attract them most from the mountains of game videos. Traditional video retrieval can only search manually through soon entering and soon retreating method, so it is a very tedious work that consuming time. This paper uses soccer video as an example, carries on content-based video semantic analysis, and realizes the semantic events detection. The purpose is to bring the convenience for the video retrieval and video automatic editing. To analyze the contents of soccer video effectively, we should detect and classify shots first, then analyze and recognize typical feature and object of court in the selective shots, finally set up effective analytical rule.Partitioning a soccer video sequence into shots or detecting shot change is one of the key techniques in video indexing. Shot change means scene content change in a video sequence. The most commonly used methods for partitioning video into shots are grey level-based methods, edge-based methods, color-histogram methods, MPEG methods, block matching methods, statistic decision methods, clustering-based methods and gradual transition detection. Soccer Video shot boundary detection is under global concern as the first step of all kinds of processing. A huge number of various methods have been proposed at previous years. The text unites different techniques to detect shot boundary through analyzing the existing research achievement.After the shot boundary detection, the next task is shot classification. The shots can be classified by the background of the soccer video which the domain color is the tone of green. Setup three threshold values for the green grass ratio, the soccer shots are classified into three classes which are long shots, medium shots, out of field or close-up shots by using this feature. It presents gold section to partition long shot and medium shot in shot classification.Feature extract is to extract various needed feature such as motorial object feature and court feature. The motorial object feature can also show the shots classification. This paper extracts the players. About the detection for players, this paper advances a series of algorithms to distill and recognize the areas of players. The problems in segmented images that a clearance often exists in a player area and the conglutination between players and lines are discussed and an appropriate method is advanced to resolve it, finally the players are recognized from the candidate areas. HMM is a statistic-based model, and is good with simulating, predicting of stochastic data. The highlight events are stochastic strongly, so this paper use HMM to describe the highlight event effectively. It has studied to use HMM to detect goal event, including how to model semantic event, how to extract features, how to train HMM and how to get the optimal state sequence using the Viterbi algorithms.With analyzing the characters of soccer video, the three hidden states of Hidden Markov Model (HMM) based on goal event is obtained, then build the HMM based on goal. To overcome insufficiency of training data, a relatively reliable measurement deduced from Baum-Welch reestimation formulas is adopted in this paper. By training samples in dataset, parameters in HMM are confirmed.Soccer video is composed with"semantic events", each of which is constructed by several"semantic shots". To analyze this semantic structure, Hidden Markov Models representing"goal highlights events"are introduced. Five features, green-grass ratio, the numbers of players, the ratio of players and the difference of the different green-grass areas ratio are constructed to be the input of HMMs as observation sequence. Using the basal algorithms of HMM to train HMMs, analyze semantic structure of soccer video.Experiments about this paper are based on windows xp and visual C++ 6.0. It is proved method presented in this paper can well detect video shots, classify video shots, extract players and recognize the highlights events.