Abstract: Corner detection is not accurate with a fixed Gaussian function parameter σ and artificially set value, so a σ parameter adaptive Harris algorithm is put forward. Gaussian function parameter σ is set in the selected region to extract preselected Harris corner. and in the process of non maximal suppression, the adaptive threshold is used to obtain the preselected angular point. The constraint criterion is adopted to choice the maximum from each σ parameter generation of the preselected angular point response function R, the effective corners are extracted and the false corners are removed. The experimental results show that this algorithm has better image corner detection performance, and can get more effective corners and less false corners.