In order to accurately evaluate the similarity of motions during daily rehabilitation training for stroke patients, this paper proposed a novel quantitative assessment method based on dynamic time warping (DTW) algorithm. Firstly, the raw accelerometer signals were preprocessed to eliminate the noise. Secondly, the similarity between the accelerometer signals and four standard task templates was calculated respectively, and then the motion was recognized based on the similarity measurements. Finally, the corresponding quantitative assessment model was used to compute the result. The clinical experimental results showed that there were significant differences in the shortest path distance (R value) of DTW between different tasks, and the classification accuracy could be up to 91% when the R value was selected as the classification feature. Additionally, with the process of rehabilitation, the R value decreased gradually, which means that the R value can be taken as the assessment index to evaluate the quality of designated tasks for stroke patients. It also indicated that the R value could be applied into the scene of automatic prescription generation and interactive gaming to determine whether it is needed to change the rehabilitation plan or adjust the game difficulty level, so as to implement the individualized rehabilitation services.