Human Security Protection Techniques Using Mean Shift Clustering

Human security is a precondition for making a good human settlement, which aims to reduce economic losses and personnel casualties by continuing hazard identification and risk management process. In order to detect the security of human settlement, the structural health monitoring and evaluation methods and the intelligent video surveillance techniques based Mean Shift clustering are studied in this thesis.The theory of Mean Shift clustering(MSC) is discussed firstly. It is proved by signal and image experiments that the selection of kernel function, kernel radius and threshold of Mean Shift can influence the performance of the MSC. When making clustering analysis with the MSC, different kernel functions will produce different results. For a kernel function, its kernel radius has an ideal range for good clustering performance. The clustering performance becomes worse when kernel radius is beyond the range. Moreover, reducing the threshold of Mean Shift is helpful to achieve good clustering performance. It is also revealed that feature extraction is an effective way to improve the efficiency and accuracy of the algorithm, which can reduce the dimensions of the classification matrix, deminish iteration times and enchance real-time performance for MSC algorithm. In addition, it is found that cluster number doesn't need to be preset using the MSC compared with k-means clustering algorithm.A structural health monitoring method based on the MSC is developed. Firstly, response signal of the ASCE structure is processed by using the orthogonal wavelet packet transform, then wavelet package energy distribution (WPED) of the signal is calculated as a feature vector. Then a group of feature vectors which obtain from different node sensors of the structure are banded as the input matrix of the MSC. On the basis of the banded feature vectors, the structural health condition is efficiently monitored by using the MSC. The experimental work shows that the feature vector banding method can make full use of the information from multi-sensors, and overcome onesidedness and uncertainty by using only one sensor. The accuracy and reliability of the structural health monitoring is improved significantly by using the banded feature vectors.A structural health evaluation method using the centroid shift is put forward in this thesis. The method firstly defineds the clustering center (which is called centroid) of the undamage condition as a reference. Then a centroid shift distance which is defined as the distance between a real-time condtion and the reference is used to assess the structural health condition. The more the centroid shifted, the more serious the structure damaged. The method can effectively evaluate the structural health condition based the centroid shift distance. A intelligent video surveillance technique based on the MSC are studied. In order to guarantee video surveillance in real time, the method extracts the histogram of video images as the feature matrix of the MSC. On this basis, changes of the protected area, such as, object motion, retentate, appearance and disappearance, are effectively detected by applying the MSC. On other hand, images of video surveillance system can be rapidly clustered by means of the MSC, the method sets a solid foundation for upgrading the speed of abnormal image identification and improving the efficiency of image retrieval in the intelligent video surveillance system.