Patients with Parkinson's disease (PD) can be divided into two subtypes based on clinical features, namely tremor-dominant (TD) and postural instability and gait difficulty (PIGD). Detection of PIGD symptoms is crucial for early diagnosis of PD and timely clinical intervention. However, patients at the early stage may not exhibit obvious motor dysfunctions during normal straight walking leading to difficulties in PD identification. Researchers have found that patients would show significant motor deteriorations in turning due to their cognition limitation. Therefore, turning detection is essential for quantitative motion analysis in the gait assessment of PD patients. In this study, we proposed a novel inertial-sensor-based algorithm for turning detection. Ten healthy young participants were enrolled in the experiment where they were required to walk along a 7-meter pathway with two 180 degree turns at their comfortable walking speed. Five inertial sensors were attached to the upper trunk, the shank and the foot of both legs. The algorithm performance was validated using an optical motion capture system for reference and two sensor combination options (upper trunk and shank sensors, upper trunk and foot sensors) were compared. The results showed that the proposed algorithm achieved accuracy over 98% for identifying the turning state of both legs. The integration of the upper trunk and foot sensors had no significant effect on the detection accuracy compared to that with the use of the upper trunk and shank sensors. Our algorithm has the potential to be implemented in the motion analysis model for complicated gait tasks, which has great potential in the early diagnosis of PIGD.