Accurate human gait phase detection is imperative in rehabilitation and robotics, forming the foundation for developing adaptive technologies and interventions. Precise gait detection enables the timing and coordination of robotic assistance or rehabilitation exercises tailored to specific gait cycle phases. In this study, a custom designed inertial measurement unit was fixated on an individual\\\'s lower back, and a peak detection method was developed for detecting heel strike and toe-off events. The method solely uses the linear acceleration data along the subject\\\'s walking direction. Two insoles with push button switches were placed in the subject\\\'s custom-designed shoes to identify instants of foot contact and lift-off. The differ-ences between the timing and peak linear accelerations of the switches were measured. Subsequently, distinct phases of gait on the right and left feet were detected, and data labeling was performed. This process was repeated for 14 trials, generating a dataset comprising the linear acceleration and angular orientation. Finally, a neural network model was designed, trained, and evaluated using this dataset. The proposed technique demonstrated a notable performance in detecting the defined phases with approximately 85% accuracy on average for all phases.

Keywords