This project consists on 3 parts:
Motion capture is about getting the orientation of every body limb or part at real time as accurate as possible. A simple MEMS IMU device*, and freely available sensor fusion algorithms are enough to get a decent result. The problem starts when you want to get the data of several devices. Most of this devices came with an i2c interface, but their address is fixed in the hardware. So one of the building blocks of Chordata is the sensing unit capable of coexisting with several “siblings” on the same bus. At the moment I have developed the “IMU-Proto” that allowed me to develop the rest of the project. It consists of a LSM9DS0 IMU, and a PCA9544A i2c multiplexer. The focus of the whole project is to reduce costs, so all the passive components on the board are through hole, passing most of the work of assembling from the industrial manufacturer to the final user while saving money in the process.
Getting the data of a lot of sensors on real time, processing it, and send it in an easy-to-read format to some client is not a simple job, so I’m developing a software from scratch to deal with it.
It is responsible for:
- Building a digital model of the physical hierarchy of sensors. Initializing the i2c communication on the Hub, and running the configuration routine on each of the sensors.
- Performing a reading on each of the sensors at the specified refresh rate.
- Correcting each sensor reading with the deviation obtained on a previous calibration process.
- Performing a sensor fusion on the corrected sensor reading, obtaining absolute orientation information in form of a quaternion.
- Sending the orientation data, together with the sensor_id and a timestamp to the client using an open protocol (such as OSC)
After several testing I discovered that using an ARM computer running linux was the best choice to host such a program, so all of the development of this part of the software has been done on C++, using a Raspberry Pi 3 as the hub. Some of the advantages of this type of hub, in comparison with simpler microcontrollers are:
- It’s not an expensive component.
- Programming and debugging is enormously simplified.
- Some of them, like the rPi3, came out of the box with all the communication peripherals needed to perform a confortable capture, with the remarkable example of the Wifi adapter.
The choice of performing the sensor fusion inside the hub is based on:
- Mayor cost of the IMU units capable of performing sensor fusion on-chip
- Mayor accuracy of the sensor fusion performed after the raw data is corrected by a previously done calibration.
- Since the bandwidth in the i2c bus generates a bottleneck in the sensor’s data acquisition, the fusion sensor processing inside the hub doesn’t add a significant overhead.
This is the less developed part at the moment. It consist on a python script running in Blender that grabs the quaternion data from OSC, and rotates the bones of a 3D armature.
Further development is planned, and the basic client software should be a Blender add-on responsible for:
- Establishing some handshake communication with the hub, checking compatibility and status.
- Communicate the status to the user.
- Act as a GUI to run the on-pose calibration procedures and start the capture.
- Display the preview of the capture on real time, and allow the user to register part of it.
- Allow an user with a basic experience on Blender to create a custom distribution of the sensors on a virtual model of the human body, export it on an structured data format (like xml) and send it to the hub.
*For the sake of simplicity here I refer to IMU device, but to be correct I should say IMU (gyroscope and accelerometer) + magnetometer