With the popularity of mobile positioning technology and location-aware devices, trajectory data plays an important role in people's lives. Qiu et al. proposed a trajectory privacy protection scheme to resist prediction attacks. However, Qiu’s scheme has insufficiency in long-term observation attacks with semantic attributes of location. In order to make up for this insufficiency, this paper proposes an improved trajectory privacy protection method based on Qiu’s scheme. The specific details are as follows. After analyzing the Qiu’s scheme, this paper proposes a new trajectory privacy protection method. Firstly, stop points in the user's trajectory are extracted. Secondly, HMM is used to calculate location predictability, adjusts privacy parameters, and generates a perturbation point set. Finally, a semantic classification tree is built based on the semantic attribute values of the perturbation point set, and the appropriate perturbation point is selected to replace the actual position. Security analysis demonstrates that the proposed scheme can effective resist long-term observation attacks with location semantics and prediction attack. Experimental results show that this scheme has the same capability as the Qiu’s scheme in terms of data utility. In summary, this proposed scheme compensates for the loop of the Qiu’s scheme in resist long-term observation attacks with semantic attributes of location and has the same data utility as the Qiu’s scheme.