Skip to Main Content
Some urban scenes exhibit periodic variations that can be relevant to visual surveillance applications. One example is the variation in the background elements, such as those caused by moving escalators, lights and scrolling advertisements. When modelled correctly, the incorporation of these periodic elements as a Markov model in a foreground detection component can improve the performance significantly. Another area where the periodicity in the scene can be used is anomaly detection. In some underground metro stations where the flow of people is periodic, deviations from this periodicity can be interpreted as abnormal movements of people. This can be achieved by using a higher-dimensional model for the underlying data structure, and mapping it to a one-dimensional signal for interpretation. This approach is tested, and the results show that abnormal behaviour can be automatically detected.