Research Of Mobile Object Recognition Based On Feature-Line Optical Flows

According to the human vision system principle, we proposed one method based on feature-line flows for agent vision to solve the limitation of the present autonomous mobile robot in the object recognition and the recognition algorithm. It extracts objects'lines or curves in the digital image sequence and some interested areas are segmented at first, and then the line or curve flows are calculated from the interested areas, the static objects or the moving objects can be distinguished by the F.O.E (Focus of Expansion) also; Finally, objects are recognized using prior knowledge which were processed by PCA. From this method, only those pixels of the feature curves (or in the interested areas) are calculated for optical flows, and it does nothing to the other areas. As a result, it saves the processing time and can run in a real-time state.For better recognition performance, we combine the feature curve flows method with a feature recognition algorithm as a second way based on object feature core which is processed by rough set. Thus, the method can detect and recognize the objects effectively. The calculating time of the arithmetic declines sharply and the vision system of mobile robot can work in a real-time state.The results of experiment show that the method performs well at both real-time and precision.