To improve the fall detection accuracy in substation, a human fall detection algorithm based on improved D-S evidence theory was proposed in this paper. Human minimum area external rectangle and vertical external rectangle were used to describe the human and the characteristics of rectangle ratio, centroid height ratio and inclination angle of object region were analyzed. The characteristic conflicts in the complex situations can be resolved by using DS evidence theory to combine the characteristic information. The basic probability assignments of three characteristics were created by using selfdefined generalized triangular fuzzy function. An improved method Dual Weighted Average Evidence (DWAE) based on Murphy algorithm was proposed to merge the different independent evidence effectively. The experimental results show that the proposed method has high detection accuracy by generating BPA reasonably and combining the evidences effectively which can be integrated in the security monitoring system of transformer substation.