Aiming at problems in authentication methods of mobile devices, such as being easy to crack or difficult to implement or high costs, this paper presented a new mobile phone acceleration sensor authentication method on human perception. To use the current widespread acceleration sensor in mobile phones to capture human motion data (typically dynamic gesture), we proposed an authentication algorithm named Window Limited Approximate Longest Common Sequence (WLALCS) based on the classical matching algorithm LCS. And we implemented an authentication system on cloud computing model. Compared with the existing gesture and accelerometer based authentication methods, this method effectively reduced the equal error rate on imitate action attack. Non-imitate attack authentication EER (Equal Error Rate) is 2%, and imitate attack authentication EER is 5%. This system is easy to deploy on any smart phone systems and does not need any additional sensors. Based on the biological characteristics, we reinforced the robustness towards mimicry attack.