The present invention relates to a method, an apparatus and a system for state of health estimation of an object sensing fusion system. Target data from a vision system and a radar system, which are used by an object sensing fusion system, are also stored in a context queue. The context queue maintains the vision and radar target data for a sequence of many frames covering a sliding window of time. The target data from the context queue are used to compute matching scores, which are indicative of how well vision targets correlate with radar targets, and vice versa. The matching scores are computed within individual frames of vision and radar data, and across a sequence of multiple frames. The matching scores are used to assess the state of health of the object sensing fusion system. If the fusion system state of health is below a certain threshold, one or more faulty sensors are identified.