Sports Types Recognition Of Sports Video With Finite-state Machine

Video retrieval content-based is an important research field of multimediadatabase, in which non-structured video frames will analyzed and processedmethodically;representative frame and dynamic features to can be extracted toform the feature index to describe shot based on the access unit shot, i.e. videosegments obtained from video streams. According to the organization of shotsand feature index, video clustering can be adopted to combine similar shots toreduce the searching scope until the expected video is found correctly. Amongthese technologies, video segmentation, representative frame and featureextraction are the key parts of those utilized in the field of video retrieval.Presently, the relevant research includes:MPEG-7, also called Multimedia Content Description Interface is astandard that describes multimedia content, in which descriptor, descriptionlanguage and scheme are defined formally to implement more efficient retrieval.JJACOB video retrieval system, in which representative frame can beextracted automatically and retrieval can be achieved by color and texturerepresentative frame description.Informedia system applied to digital video library designed by CarniegeMellon University, in which speech recognition, video analysis and text retrievaltechnologies and which will support 2000-hour retrieval of vide broadcast toimplement full-content search and retrieval based on knowledge.Against to other video types, there are some unique characteristics tosports video. Usually, a sports program occurs in some specific area and thereare some cameras for tracing and rich movement information and contentframework defined precisely are in it. Therefore,the determination methodbased on finite-state machine is used to classify and recognize sports video canbe found.With more and more video applications appearing in Internet, we needmore efficient methods to solve video management and retrieval, especiallysemantic object. In this paper, the semantic object detection and segmentation isthe first step to video retrieval at the semantic level;an efficient algorithm tosegment objects such as athletes and field in sports video by analyzing theinformation of visual and motional features is proposed, in which Konradalgorithm and shots similarity is adopted to detect the sports field, he result isverified by dominant color calculation histogram and the information of field isapplied to detect and segment the motion areas in the sports field. Theexperimental results show that it is satisfied and it is significant to video analysisand retrieval that accurate segmentation of sports and athletes areas.People wish to find their favourite programs from mass videos,;here, themodel is formed for different sports matches by finite-state machine and then thecorresponding template is called to implement quick retrieval according to users'requirements.An important step in the process of video structure parsing is that ofvideos are segmented into individual scenes, any of which consists of a seriesof consecutive shots. Then these shots are grouped together because they arefilmed in the same location or they share some thematic content. The processof detecting these video scenes is analogous to paragraphing in text documentparsing, but it requires a higher level of content analysis. In contrast, actually,shots are physical basic layers in video, whose boundaries are determined byediting points or where the camera switches on or off. Fortunately, analogousto words or sentences in text documents, shots are a good choice as the basicunit for video content indexing, and they provide the basis for constructing atable of contents for video. Shot boundary detection algorithms that rely onlyon visual information contained in the video frames can segment the video intoframes with similar visual contents. Grouping the shots into semanticallymeaningful segments such as stories, however, usually is not possible withoutincorporating information from the other components of the video. However, itis not feasible that video segmentations with rich semantic information withoutother video components. To do so, multimedia processing methods areapplicable including video structure, text, speech and language, etc.The comparison is done to several algorithms used to construct FSM.Experiments show that Baum-Welch the quality of algorithm is over other twoalgorithms. Therefore, FSM is adopted to train data. To construct the FSM,input video series is segmented into shots, and then the estimation to globalmotion parameters will be achieved by the algorithm of Konrad and Fisherlinear determination;motion object will be extracted and the scene isconstructed through similarity measurement of the neighbor frames. Duringthe process, dominant color calculation histogram and accumulative histogramis introduced to trace and extract the dominant color that lasts long.Based on the semantic segmentation of video sequences, differentsemantic segmentation with diverse semantic concepts will be reflected on thestate chains of finite-state machine to be trained by video sequences based onFSM.The construction of FSM based on four most typically sports matches isimplemented in this paper and its quality is evaluated from the accuracy andfullness.The maximum and optimal match is utilized during the video retrievaland recognition and tests show that better speed and accurateness can beachieved. In the paper, the FSM-based video retrieval system is designed byWindows 2000 OS and developed by VC++Builder 6.0, which can achievebetter a retrieval function. The simulation results show that the system is reliableand steady.