Construction Crane Hand Signal Recognition Using Wearable EMG-IMU Sensor Fusion
Effective communication between signal persons and operators is critical to preventing construction crane fatalities. However, traditional visual hand signals are frequently compromised by line-of-sight obstructions, and radio communication suffers from high noise in construction environments. To overcome these constraints, this study proposes a non-line-of-sight communication framework that uses a wearable electromyography (EMG) and inertial measurement unit (IMU) fusion system. A modality-aware four-branch 1D convolutional neural network was developed to classify 16 standardized crane hand signals from 12 participants using a single forearm armband. By independently processing the EMG (time-domain and power spectral density) and IMU (accelerometer and gyroscope) streams, the proposed architecture effectively integrated muscle activation patterns and arm kinematics. This is the first application of EMG-IMU fusion to construction crane hand signal recognition, with a branch design informed by a signal-space characterization of the hand signals. The model achieved 99.69% accuracy under subject-dependent conditions, whereas the subject-independent evaluation yielded 79.30% accuracy. Modality ablation confirmed the necessity of sensor fusion because the fusion model outperformed the IMU-only and EMG-only models by 9.35 and 44.82 percentage points, respectively. Furthermore, unsegmented continuous sequence testing achieved an average accuracy of 86.1%. This study demonstrates the feasibility of wearable EMG-IMU fusion as a non-visual alternative for crane guidance, laying the foundation for safer and more reliable communication during crane operations.