This paper presents an agent-based architecture designed to functionally combine data from an homogeneous network of sensors fro tracking purposes. The system has been developed in a video surveillance context to detect, classify and track moving objects in a scene of interest. Although single camera systems could perform the tasks outlined above, they wouldn't be able to deal with topologically complex environments such as corridor; corners and indoor locations in general. The multi-sensor approach has been used to overcome these problems, nevertheless issues arise such as data fusion, synchronization and camera calibration. The sensor fusion approach here proposed uses autonomous software agents to negotiate the combination of data and the fusion is carried out by appropriate signal processing algorithms. The system has been tested with indoor video sequences to show system's capability of preserving identity and of correct trajectory estimation of tracked object.