Skip to Main Content
A multi-cue method is put forward to solve the problem of multiple objects tracking under family environmental service. The cues mainly include: object detection, object prediction, and object tracking. Motion History Image is used to detect foreground and the connected component analysis is adopted to establish the target measurements. Kalman filter is presented to predict the old objects, if two of the predicted objects are closely enough in the space then merge the both. After prediction merging, merging prediction matches with the measurement according to similarity in location. If the measurements are close enough, then object tracking by Mean-shift is introduced, then the tracking and the matched measurement are treated by their similarity in features with that of previous frame. If the measurements cann't match with the prediction, the measurements is corresponding to a new object. At last, the prediction results are considered. If the prediction cann't match with none of the measurements, then object match based on Mean-shift is used to locate the object.