In localization problems of autonomous robots， if accurate environmental features can be extracted from the scanning data of laser radar， the matching speed in simultaneous localization and mapping （SLAM） will be greatly improved. At present， most approaches for feature extraction adopt the iteration strategy， which have high computational complexity. To overcome these drawbacks， a new algorithm is proposed to extract corner feature.In the method， iteration is avoided， and on the premise of ensuring the accuracy， the least square method is replaced by two-point fitting line by positioning the corner points to modify the segmentation result. First， the length and angle of the scanning points obtained from the laser radar are used to calculate the slope difference of the adjacent points for the initial segmentation of the point set. Then， after calculating the slopes of the line segments corresponding to each point set， the point set is merged to solve the over-segmentation problem. Finally，the corner feature is located and extracted by calculating the intersection point of two adjacent lines. Experimental results show that the developed algorithm can extract corner features from the scanning data accurately and has better position accuracy and computational efficiency.