Inertial Motion Tracking Dataset, IMTD

The project research team successfully established China's first Inertial Motion Tracking Dataset (IMTD), which can be widely used for artificial intelligence model training in fields such as satellite-free navigation, unmanned driving, and wearable devices. Based on the IMTD dataset, the motion tracking method proposed by Wang Yifeng, Zhao Yi, and others breaks through the limitations of traditional motion tracking and positioning technologies such as inertia, optics, GPS, and carrier phase. It provides new technical solutions for wearable devices, human-computer interaction, aerospace, and other fields, and has broad market application prospects.

The motion tracking technology based on inertial sensors has extensive application value in fields such as aerospace, virtual reality, human-machine interaction, and unmanned driving. However, due to the accuracy of inertial sensors, the technical difficulty of 3D motion trajectory restoration is high, and currently, the academic community has not been able to break through the arbitrary trajectory reconstruction technology based on a single inertial sensor. Some studies attempt to use deep learning models to predict motion trajectories, but three-dimensional motion trajectories have a large number of parameters to be predicted, and the limited training samples are a drop in the bucket compared to the massive number of parameters to be predicted. To this end, the research team has established a manifold deep learning scheme, which realizes the low-dimensional representation of high-dimensional trajectory manifolds through the Geometric modeling library, thus reducing the learning cost of the deep learning method and ultimately realizing arbitrary motion tracking and trajectory reconstruction based on a single inertial sensor.