Smart Full-Body Fitness Tracking

Figure 1. An on-screen robot displays a video of your workout so you can follow along. (Image: Bosch Sensortec)

I’m guessing that when you think about applications for AI, physical fitness is not the first thing that comes to mind. But it is actually one of the more common applications. The most popular fitness sensors use AI to keep track of how many steps you’ve taken, track your progress, and integrate that with health information like heartrate or even blood oxygen level. But AI can do much more than that.

A case in point, is the new Smart Connected Sensors platform from Bosch Sensortec, Reutlingen, Germany. With it, you can track more than just steps. You can program complex whole-body movements and accurately track them during physical workouts or while you are going through a rehabilitation or physical therapy regimen.

The system can recognize movements and count repetitions. Beyond that, however, it provides real-time feedback on the quality of how well the movements are being performed. This enables users to learn, adapt, and improve the quality and consistency of each repetition of all the different movements that make up part of the routine.

When using this platform for a fitness tracker, AI-powered software watches how you do a certain movement. For example, you can teach the system what you want to do by positioning your arm at a right angle. And teach it that you want to get to 20 per day in a certain amount of time. The software saves that information, watches you as you do your exercises and tells you how well you did. Perhaps one day, not so good: “you need to make sure you do the right angle better and faster.” It keeps monitoring your performance and shares a record of your progress.

You can create a robot avatar on a tablet screen, for example, and calibrate it by teaching it the precise movements you want to achieve. The on-screen robot will then display a video of your workout so you can follow along. As you go through the routine the image gives you real-time visual feedback by changing the color of stripes located at each of the sensor nodes in the display so you can see exactly where you’re going wrong. When you’re finished, you get an overall score for your completed workout.

Figure 1 is an example of the sensor platform in action. The sensors on the limbs capture a wide range of movements, including lifting arms and legs, twisting, turning, even doing a little shimmy. And it is even sensitive enough to follow the rotating of the palm of your hand.

Hardware

Figure 2. The Bosch Sensortec BHI380 smart sensor is a programmable IMU. (Image: Bosch Sensortec)

Central to the system is the Bosch Sensortec BHI380 smart sensor, which is a programmable inertial measurement unit (IMU). According to an NIH Study, “… the development of IMUs that meet needs clearly expressed by clinicians will provide them with new opportunities in their clinical practice. Like several other studies, our work highlights the importance of a user-centered design approach and of a close collaboration between engineers, researchers and end-users to promote IMUs acceptance.”¹

Among the many advanced features of the BHI380 is the ability to collect data from up to eight sensors attached to different parts of the body and synchronize them to create an overall picture of body movements.

The hardware components embedded in the device, which enable this functionality, include a 16-bit 3-axis accelerometer, a 16-bit 3-axis gyroscope, a 32-bit ARC^™ EM4 CPU with Floating Point Unit (FPU), 256 kB on-chip SRAM, 144 kB on-chip ROM. Interface options include several GPIOs, and multiple I2C and SPI interfaces. The accelerometer and gyroscope provide the precise tracking and position feedback.

Software

Figure 3. It’s especially important when the system is used for physical therapy to precisely track your movements. (Image: Bosch Sensortec)

The software comprises an Integrated Event-Driven Software Framework built on top of an OPENRTOS (real-time operating system kernel for embedded devices), an integrated Bosch Sensortec BSX sensor fusion software library with autocalibration algorithms, 3D device orientation, gravity vector, and ultra-low-power algorithms such as step counter, tap detection, gesture detection, and activity recognition. Additionally, it includes self-learning AI software suitable for various fitness tracking and custom motion patterns.

The BSX sensor fusion software is designed to integrate data from multiple sensors to provide accurate and reliable tracking of the user’s movements. It includes algorithms for sensor data processing, calibration, and fusion, as well as APIs and tools for developers to integrate it into their applications. The available fusion algorithms can effectively combine sensor signals, including acceleration, angular rate, and magnetic field strength. These algorithms process the sensor data and produce useful information such as orientation, linear acceleration, tilt-compensated heading, and gravity.

The System

The BHI380 sensors can be used as nodes on a multi-node system, such as body area network applications. A hardware reference design includes a customized version of the firmware with sensor data synchronization software. The wearable reference design can be straightforwardly attached to any part of the body without having to handle cable clutter, an external camera or dedicated bodysuits. The wearable reference design utilizes the Bosch Sensortec Smart Connected Sensors (SCS) platform, for communication among the different sensors via Bluetooth.

The number and placement of sensors depends on the application, the type of gestures, and the required accuracy. For applications requiring precise movement analysis for multiple body parts, such as rehabilitation or high-end sports, a multi-node sensor solution is necessary. The SCS supports up to eight sensor nodes, which can be flexibly placed on any body part. Integrated time-synchronization algorithms, which are crucial for coordinating the data from the various body nodes, are included in order to construct a meaningful whole-body model.

The network uses Bluetooth LE to connect the nodes and synchronization software to create a coordinated image. The nodes are connected in a star topology with one central node and up to seven leaf nodes. Since each node has the same sensing and compute capabilities, any one can be chosen as the central node of the star. Once you’ve chosen the central node, it will be the one that collects and processes all of the data and organizes the network.

Time synchronization is of course vital to the operation of the network, and never more so than when the system is being used for physical therapy. There it’s especially important to precisely track even tiny movements so the therapist can accurately track your progress.

The BHI380 can also connect to other sensors through secondary interfaces. It integrates and synchronizes both external and internal sensor data. Common external sensors include barometric pressure sensors and magnetometers. It can connect to additional sensors via SPI or I2C, such as proximity sensors, light sensors, heart rate sensors, or even additional IMUs.

Barometric pressure sensors enable altitude to be accurately determined. The BMP585 measures a change in height of just a few centimeters, so it can detect movements in fitness training down to the level of individual pull-ups or push-ups. And the BMM350 magnetometer provides information about angles and rotations.

With AI, we’re clearly on the way to the next level of fitness tracking.

This article was written by Ed Brown, editor of Sensor Technology. For more information, go here .

Reference

Clinicians’ perspectives on inertial measurement units in clinical practice PLoS One. 2020; 15 (11): e0241922. Published online 2020 Nov.13. doi: 10.1371/journal.pone.0241922