Event detection with vector similarity based on fourier transformation

Event detection through sensors data recording human activities is an aspect to learn human behaviors. In this paper, counted numbers from a sensor installed on a building entrance recording the number of people entering the building, will be processed to find the anomaly time interval when there are more people going through the entrance, which is viewed as event. An approach is adopted having two steps: first, the counted numbers over time is processed by Fourier Transformation and we get the parameter of a vector (ReX[k], ImX[k]) representing kth point in the data set; second, the vectors of (ReX[k], ImX[k]) are classified by KNN algorithm in two dimensions, categorizing the data in the same time interval in 70 days and the data in 48 intervals in one day. The results show that the proposed method works well.