With the original DTW algorithm, it is required to calculate each point-to-point accumulated distance completely in order to infer backward from the end point to the initial point for locating the shortest path. To facilitate real-time comparison of movements during rehabilitation, a modified algorithm is proposed in this study. This paper introduces a DTW-based fast searching heuristic suitable for comparing rehabilitation movements to help reinforce the calculation process required to locate the shortest path and to reduce the amount of calculation involved. The algorithm is also the basis for the development of the frozen shoulder rehabilitation system. Combining Kinect technology and using Processing programming language along with the SimpleOpenNI function library, this system is capable of detecting the skeleton, calculating the shoulder and elbow angles, and quickly comparing movements by means of the introduced modified algorithm. Results of the study show that the total amount of calculation involved with the modified algorithm is approximately 2.37% of the original one and the located shortest path also highly resembles that obtained with the original algorithm. The system provides measurement for correcting rehabilitation movements in real time and records movements during rehabilitation for patients' and healthcare pro- fessionals' reference via low-priced consumer equipment.