**4. Discussion**

The gait experiment with the two-IMU network system showed promising results in measuring three-dimensional joint angles. One of the main contributions of this paper is to provide a procedure to calculate the three-dimensional relative angles of the knee joint with respect to the hip joint using a pair of IMUs. It seems that the IMU-based method has lower accuracy compared to an optical camera-based method, such as the Vicon system [24]. However, the preliminary experimental results showed similar waveforms of hip and knee angles with comparable peak heights [24]. Unknown initial joint angles and error accumulation in the integral value of the gyroscope were still a significant problem. Further, there were some minor differences when observing the results, such as some abrupt changes at peaks or valleys during motion. There could be several reasons for the differences: human error or sensitivity of the sensors. Human error could be from mounting the sensors. If there was some error from the mounting, it did not appear to be significant.

In the identification of the characteristics of human gait, the study proposes using phase difference angle patterns. Figure 7 and Table 1 show evidence of the phase difference angle changes between the hip and joint movements during hemiplegic walking. Table 1 shows that the phase difference angles in all the directions were decreasing consistently.

It is known that a stroke commonly results in hemiplegic gait cycle impairments that are associated with decreased hip and knee coordination and limited leg muscle strength. These impairments often result in biomechanical changes during walking. The figures reflect the effects of reduced control of the leg swing that results in the decreased phase difference angles. The three-dimensional phase difference angle patterns are suitable for training machine learning methods to detect gait changes during walking [25,26].
