Intelligent Human Motion™ (IHM) is a targeted application of Edge ML for real-time motion and activity recognition on edge devices. IHM works similar to image and speech recognition: sensor signal patterns are analyzed and compared in real-time to trained models to determine a user’s context such as motion and posture. This contextual data is obtained from commercially available sensor data from smartphones, wearables, and AR/MR/VR headsets, but has also been cross-compiled for custom hardware platforms.
IHM algorithms maximize classification accuracy while minimizing computational complexity. IHM comes with a database of pre-trained motion models ready to use for your application. Additionally, with MLOPs you can quickly create your own motion models. Training can occur offline or directly on your edge device.
· Specialized Models: IHM runs multiple models simultaneously. Each model represents a particular motion type and each model can be computed in parallel. Because each model is only responsible for a single classifier it is small in both memory footprint and computational load.
· Stacked Models: Unlike Deep Neural Networks, a new motion classifier can be added to the model set in real-time without affecting the current models or requiring a network to be retrained for the added motion type output.
· Minimal Data Requirements: Training a motion model requires very little data. Often less than a minute of data is needed to train a model capable of 95% accuracy.
· Attitude Independent: The IHM algorithm does not require the user to hold or wear the device in a particular location or orientation. Specialized filters are used to pre-process the data to reduce the effects of device orientation.
IHM is available today and ready for integration in your product. We have a mobile application in closed beta for demonstration. Please get in touch with our business team to apply for a product demonstration today.
