Skip to Main Content
Object detection is a fundamental aspect in surveillance systems. Although several works aimed at detecting objects in video sequences have been reported, many are not tolerant to dynamic background or require complex computation in addition to manual parameter adjustments. This paper proposes an adaptive object detection method to work in dynamic backgrounds without human intervention. The proposed method is based on a neural-fuzzy model. The neural stage, based on a one-to-one self-organizing map (SOM) architecture, deals with the dynamic background for object detection as well as shadow elimination. The fuzzy inference Sugeno system mimics human behavior to automatically adjust the main parameters involved in the SOM detection model, making the system independent of the scenario. Results of the model over real video scenes show its robustness. These findings are comparable to the results obtained with human intervention to define the parameters of the model. A quantitative comparison with methods reported in the literature is also provided to show the performance of the system.