Abstract: Gait event detection (GED) plays an important role in clinical gait analysis and rehabilitation. Real-time detection of temporal gait features within the gait cycle is essential for closed-loop control of wearable assistive devices and neuroprostheses. Machine-learning algorithms used in these systems require training datasets of real-world walking, which are currently collected using insole footswitches. However, insole footswitches suffer from performance and reliability issues. This study evaluates the use of inertial measurement units (IMUs) as an alternative for detecting foot-ground contact events during gait. The detection performance of five IMU placement locations and four rule-based detection algorithms was assessed in laboratory and real-world settings, using force plates and footswitches as ground-truth reference. In the laboratory, the heel-mounted IMU combined with a vertical jerk–based algorithm achieved an F1-score of 97.6% and a mean timing error of –1.5 ± 1.5 ms relative to force plates. Although insole footswitches achieved a comparable F1-score (97.2%), they showed significantly larger timing errors (17.7 ± 11.7 ms, p <
0.001). In the real-world environments, the heel-mounted IMU combined with a zero-velocity update–based detection algorithm achieved a mean timing error of –5.0 ± 23.4 ms relative to the footswitches. These findings show that a single heel-mounted IMU combined with a rule-based detection algorithm offers a practical and accurate alternative to insole footswitches for collecting gait datasets in real-world environments.