Soccer Video Semantic Analysis Based On Multi-modal

With the development of technology, people obtain information more quickly, and the amount of information is increasing. Nowadays, people share all of the information in the world. People are taking a firm step to the information society. There is a hot topic that how to get interested information from all of the information. In order to meet the requirements, information retrieval is born. Nowadays, the traditional text-based retrieval has not been able to satisfy the requirement of people. At present, people is no longer faced problems by the lack of multimedia content, and how to find the semantic content is a hot spot issue in information retrieval. For soccer videos, there is only a few minutes highlight content in 90 minutes'football match. How to search highlight shots based on the semantic information becomes the problem of today's research. According to the features of soccer video, a method of detection highlight soccer video is proposed. This method is mainly divided into the following step:Using the dual threshold of hsv histogram, we split the video camera into shots, and select 3 key frames in each shot for analyzing the data in the video. The soccer shot types can be divided into long shots, in-field medium shots and close-up shots or out of field shots. The paper propose a method to classify the shots, which is used of the grass rate, the grass rate of the below 10% in the picture,the second main color and the most area of person in the field. First, use the grass rate to find out part of long shots. Then, use the grass rate of the below 10% in the picture to check out close-up shots or out of field shots. Finally, use the second main color and the most area of person in the field to classify another long shots and in-field medium shots.We use Hough transformation to extract lines in soccer video. First, separate field dominant from background dominant used the range of grass green values and Hough transformation. Then, deal the pixels in the field dominant through gray stretch, edge detection, binary image, morphological processing, thinning to get the white pixel in the picture. Finally, use Hough transformation again to check out white pixels on lines in field dominant and remove noise part.When a wonderful event occurs, the comments sound excited for the audio in soccer video. And lead to short time average energy in high level. Therefore, set reasonable threshold to check out the short time average energy in order to check out the excited part.Finally, to check the event of goal is as follows: after long shots with lines is close-up shots or out of field shots, then long shots with lines shown in the video, as well as there is exciting sound in the audio.Summarizing the content above, a simulation system of shot event detection for soccer video on the platform of Visual c + + is devised. The experimental results show that this scheme can effectively detect the highlight events in soccer videos.