Abstract: To detect outdoor shadow edges from live videos with moving viewpoints,an online algorithm of detecting outdoor shadow edges from live videos is proposed.Firstly,we divide shadow edges into two parts,i.e.the original shadow edges tracked by optical flow from the previous frame and the new shadow edges caused by the camera movements.As the tracking cumulative errors are inevitably generated,the initial tracking results are firstly refined by incorporating optimization algorithms based on the confidence of Canny edges and the homography matrix.When detecting new shadow edges,we observe that both the tracked and new shadow edges in a same frame have similar lighting conditions and thus similar shadow features.Hence,we extract the features from optimized tracking shadow edges,and then match the new edges with the set of features and pick out the new shadow edges.Finally,the gradient coherence is introduced as a constraint to line the breakpoints of both tracked and new shadow edges.For live outdoor videos under moving viewpoints captured by handhold cameras,the comparison results with the optical flow tracking and ground truth show that the proposed algorithm can completely detect the shadow edges with higher accuracy.