The research team of Yujiao Hao in Canada has designed a new cross-modal inertial measurement simulator, CROMOSim, based on deep learning, with which to simulate high-fidelity virtual IMU (inertial measurement unit) data from motion capture systems or monocular RGB cameras.

CROMOSim consists of three functional modules: an input data processing module that extracts global human motion sequences from source data, a human model that fully represents the extracted sequences and can be sampled from any body position, and a simulator module that converts noisy motion sequences into high-fidelity IMU readings.

The team conducted two sets of experiments to evaluate the performance of CROMOSim, and the results showed that CROMOSim was 6.7% more accurate than the benchmark method in human activity recognition.