Abnormal Objects Recognition In Video Based On Data Mining

The current video monitoring system only plays a function as monitoring and storage function, many important and abnormal events in video are found by manual operating, which is laborious and always miss many important information. Why is the monitoring system not intelligent? The essential reason is that there is a wide gap between low-level image information understood by computer and high-level semantic information understood by human. The important factor of achieving intelligent monitoring system is to across the gap. Data mining can find potential, unknown and valuable information from huge amounts of data. Applying data mining to analysis and recognition of video data can convert low-level image information to high-level semantic information, and which is a new research direction of data mining.A target whose movement feature, expression feature and trajectory feature are different from normal behavior is a abnormal target. The recognition of abnormal target is to establish a description model for the object by using a set of effective features. There may be a huge number of moving targets in the same time in video monitoring, so extracting the abnormal objects from many goals involves to classification by model of action(normal and abnormal). The first step is to segment objects and extract feature form video automatically. The second step is to select the suitable algorithm for mining features and construct the classifier at the same time. The last step is to classify the objects in video by using the classifier. The video monitoring system with that function needs many technologies, including object detection and segmentation, target tracking, feature extraction and data mining.The research direction of this paper is to apply data mining to the recognition of abnormal in video monitoring system. The main research contents are as follows:(1)Doing research and analysis the classification algorithms that we always use in data mining, including the decision tree, the bayes, the neural network, k-Nearest neighbor and classification based on association rules. And the same time compares the classification accuracy by the experiment. Combination the evaluation criteria of classification algorithm and the function the intelligent monitoring system needed, selecting the classification based on association rules as the main algorithm of data mining.(2) The algorithm of classification base on association rules is researched and the adverse consequence which is caused by the value of minimum support and minimum confidence is analyzed.(3)Because the traditional associative classification algorithm is sensitive to the minimum support and minimum confidence, we use the simulated annealing algorithm to optimize the minimum support and minimum confidence and propose the new algorithm SC. SC overcomes the disadvantage that the minimum support and minimum confidence of traditional algorithm causes a low accuracy, and make the setting of value can combine well with the problem size, so that the accuracy of classification can be best. At the last,compare the accuracy of traditional algorithm, HC and SC by experiment.(4) The intelligent monitoring system which has a function of recognition abnormal targets by using data mining technology SC is designed. This system is composed by data mining system and user system. For the video user selected, the system can segment objects, extract features and associative classification. And at the same time, the system can storage the information of abnormal objects in the database, so that the user can brows the information easily.