According to the problems of low-precision and less real-time in Harris feature point matching, proposes an improved algorithm of Harris feature point matching. Firstly, Harris feature point extracting algorithm is improved so that it can reduce false feature point extracting. The initial feature point pairs are extracted with the method of improved bidirectional greatest normalized correlative coefficient, and the false feature point pairs are rejected by the improved random sample consensus algorithm to realize precision matching. The results indicate that the improved algorithm not only improve the precision of feature point matching but also improve the efficiency of feature point matching greatly.