Abstract: Aiming at the problem that it is complex to determine the wheelset profile in the online wheelset profiles detection system, a novle algorithm on wheelset profile extraction is proposed. Firstly, six types of typical wheelset profiles are analyzed. Then, the DBSCAN method is used to culster the profile data and an improved Huff transform is employed to determine the effective area of the profile. Secondly, a curve fit is performed to obtain the reference point of wheelset rim based on the curvature entropy and the least square methods. Thirdly, the complete wheel profile can be acquired after the operations of coordinate rotation and translation. Finally, the feasibility of the proposed algorithm is verified through field experiments. The results indicate that the precision of wheel profiles acquired from the online detection system is higher than that of the geometrical profiles measured manually and the proposed algorithm can satisfy the requirement of profile calculation of the online detection system.