The Magic Wand Example

Introduction

The magic wand example shows how to detect gestures using an accelerometer. Different types of accelerometers are easily available:

• The ADXL345 is a tree axis digital accelerometer by Analog Devices
  I wrote a pure MircoPython driver for the ADXL345 available at https://github.com/uraich/IoT-Course/tree/master/exercises/solutions/exercise_16/adxl345.
  
  
• The MPU6050 is a six axis device (Gyro and Accelerometer) from Ivensense. It is described by its datasheet and the register description document.
  Jeff Rowberg has written a very complete MPU6050 driver for the Arduino in C++, which includes access to its Digital Motion Processor, whose description cannot be found anywhere on the WEB. I ported this driver to MicroPython together with a few example programs drafted after Jeff's C++ examples: https://github.com/uraich/IoT-Course/tree/master/exercises/solutions/exercise_16/mpu6050.
  
  .
  
  
• The MPU9250 is very similar to he MPU6050 but has a magnetometer in addition. Here is its data sheet and the register map. The MPU9250/6500 has the necessary pins to be accessed either through I2C or SPI.
  Attention: I bought an "MPU9250" board marked MPU92.65 with a chip marked MP92. This chip is either only a MPU6500 or it is faulty. It is probably a fake. I never managed to access the AK8963 magnetometer, which is part of a real MPU9250.
  

The MPU6050 and the MPU9250 gave hardware offset registers which allow calibrating the device. When the device is horizontally mounted and stationary (not moving) we expect the values:

• 0,0,1g for the measured acceleration in x, y and z, z pointing in the direction of gravity.
• and 0,0,0 for the gyroscope

The measured values may however differ. Setting the hardware offsets for acceleration and gyroscope permit to correct for the offset. The application note;
http://www.digikey.com/en/pdf/i/invensense/mpu-hardware-offset-registers describes these registers.

• The LIS2DH MEMS digital output motion sensor from ST Microelectronics is an ultra low power high performance 3 axis accelerometer.

Hardware

I mounted the accelerometer onto a WeMos D1 prototype board. With the triple base and the CPU this makes up for a stable assembly that can easily be moved without cable contact problems.

Hardware connections

The MPU6050 is connected to the WeMos D1 bus as follows:

MPU6050	WeMos D1 bus	GPIO
Vcc	5V
GND	GND
SCL	D1	22
SDA	D2	21
INT	D0	26

XDA and XCL are foreseen to control an external I2C bus, These pins are currently not used. AD0 allows to modify the MPU6050 I2C address and is also not used.

The connections of the MPU92/63 are similar

MPU92/63	!WeMos	GPIO	I2C	SPI
Vcc	5V
GND	GND
SCL	D5	GPIO 18	SCL	SCK
SDA	D7	GPIO 23	SDA	MOSI
AD0	D6	GPIO 19	Alternate module address	MISO
INT	D1	GPIO22
NCS	D0			CS
FSYNC	GND

The ADXL345 is connected as follows:

ADXL345	WeMos D1 bus	GPIO
Vcc	5V
GND	GND
SCL	D1	22
SDA	D2	21
INT1	D6	19
INT2	D5	18

While the MPU6050 and the ADXL345 use the I2C bus, I connected the LIS3DH to be used with the Serial Peripheral Interface (SPI) or I2C as follows:

LIS3DH	SPI pins	WeMos D1 bus	GPIO
Vcc		3.3V
GND		GND
SCL	SCK	D5	GPIO 18
SDA	MOSI	D7	GPIO 23
SDO	MISO	D6	GPIO 19
CS	CS	D0	GPIO 26
INT1		D1	GPIO 22
INT2		D2	GPIO 21
ADC1		D3	GPIO 16 on ESP32 WROOM, GPIO 25 on ESP32 WROVER-B model T7 V1.5
ADC2		D4	GPIO 17 on ESP32 WROOM, GPIO 27 on ESP32 WROVER-B model T7 V1.5
ADC3		D8	GPIO 5

Remark: The LISDH breakout board has 3 SMD jumpers marked "0" on the photo above. These tie ADC channel zero and two to GND and channel one to Vcc. These jumpers must be removed if you want to use the ADC. The ADC however is of limited use as it only works in a range of ~ 800mV to 1.6V.

Providing a training data set

In order to provide a training data set, we must be able to record gestures. We must therefore provide a program that recognizes the start and the end of a gesture (movement and inactivity detection) and which records the accelerometer data of the gesture onto a file. Gesture detection is of course also needed to be able to feed the accelerometer data into the trained model, which will then recognize the gesture.

-- Uli Raich - 2022-02-02

Comments