Object Tracking method and apparatus for a non-overlapping-sensor network

An object tracking system for a non-overlapping sensor network, comprising:

• a plurality of sensors, at least an entrance or exit being marked out for each of said plurality of sensors;
• a training-phase processing module that obtains a plurality of measured data as training samples through each of said plurality of sensors, and records all leaving events during a past period in a training sample space for each entering event at each entrance or exit of each of said plurality of sensors;
• a characteristic function estimating and updating module, with existing samples in said training sample space and through an automatic learning method, estimating at least three characteristic functions related to an object to be tracked, including sensor spatial correlation function, moving time difference function and appearance similarity function;
• and
• a detection-phase tracking module that uses estimated said three characteristic functions as basis for tracking said object and linking relationship of said object;
• wherein said automatic learning method is a recursive learning strategy;
• and
• wherein said recursive learning strategy includes:
• (a) for each entering event occurring at each entrance or exit of each of said plurality of sensors, recording all leaving events during a past period in a training sample space;
• (b) with existing samples in said training sample space, estimating an entrance or exit correlation probability function, a moving time difference probability function and an appearance similarity difference probability function;
• (c) observing said appearance similarity difference probability function and removing data belonging to outliers statistically;
• (d) with remaining data, updating said moving time difference probability function and said appearance similarity difference probability function;
• and
• repeating steps (c) and (d) until said moving time difference probability function converges.