The Application Of Statistical Learning Based Artificial Intelligence In Digital Games And Digital Entertainment

With Rapid development of computer and multimedia technologies, digital games and digital entertainment have become a hotspot of computer technology, and the industry of digital entertainment has become more and more prosper. This thesis focuses on the artificial intelligence application based on statistical learning which applies to the area of digital entertainment and get some achievements.After introducing the development and revolution of the AI using in digital entertainment, some mainstream AI technologies related with the game design were listed and the relationship between AI technology and game was discussed. Afterwards, the opinion that the AI in the game is context related was pointed out. The machine learning was considered a special and important method in this area.Traditional statistical learning studies the theory which based the on infinite samples. On the contrary, the Support Vector Machine(SVM) is a machine learning algorithm which is based on small quantity samples. Using SVM can make the Structural Risk Minimization. The method based on SVM which apply to the digital game and entertainment was pointed out.Computer Go is a difficult problem in the nowadays AI area. Depending on the model by which human player plays the game, a general design method used by computer Go game was mentioned. Some well-known computer Go programs were listed and their characteristics were discussed. Afterwards, the method based on the statistical learning was pointed out. A Common Fate Graph representation method was introduced. Experiment results prove that the method is reasonable and more important, it presented us a good idea.The recognition of multimedia content has become an important technology in digital entertainment. This thesis presents an audio-visual hierarchical model to detect explosion scenes from Mpeg stream based on compressed features: first, we use a coarse SVM to discriminate explosion and explosion-like audio from others, then several fine-grained SVMs are used to determine explosion audio from explosion-likeone. From these coarse to fine-grained SVMs, the audio explosion candidates are selected out. Because most explosion scenes have obvious visual change, the corresponding video is checked to get the final result in the end. The experiment result show the method is very useful.