Mostra la font per a Classificació de gestos emprant la placa IoT-02

[[Image:00gestos.png|center|thumb|Acceleròmetre / giroscopi MPU-6050 connectat a la placa IoT02]]

= Instal·lació d'eines d'aprenentatge automàtic (''machine learning'') i ''tensorflow'' emprant Python =
Es pot fer que totes les eines necessàries funcionin sense necessitat de crear un entorn virtual. Si teniu altres aplicacions funcionant en Python potser no us interessa haver de substituir paquets instal·lats amb ''pip'' per versions prèvies, per a poder fer compatibles totes les eines.

Una eina molt estesa per a crear entorns virtuals en Python és Anaconda. Si no teniu instal·lat [https://www.anaconda.com/ Anaconda], podeu fer la instal·lació.

En cas de tenir actiu ''conda'', desactiveu-ho:
 conda deactivate

Feu un entorn per a treballar amb ''tensorflow'':  
 conda create -n ml tensorflow
 conda activate ml
 pip install everywhereml
per a sortir de l'entorn conda:
 conda deactivate

= Captació de les dades =
Connecteu la placa MPU-6050 a la [[Placa IoT-02]]. Pugeu el codi [https://raw.githubusercontent.com/jordibinefa/IoT-02/master/codes/IoT-02_mpu6050_dataForwarder/IoT-02_mpu6050_dataForwarder.ino IoT-02_mpu6050_dataForwarder.ino]

Canvieu el port de comunicacions (a Linux '''/dev/ttyUSB0''') d'aquest tros del codi [https://github.com/jordibinefa/IoT-02/blob/master/codes/python/machineLearning/collector02.py collector02.py] per adaptar-lo al vostre port sèrie (per exemple '''COM3''' a Windows):
    imu_collector = SerialCollector(
        port=''''/dev/ttyUSB0'''', 
        baud=115200, 
        start_of_frame='IMU:', 
        feature_names=['ax', 'ay', 'az', 'gx', 'gy', 'gz']
    )
    imu_dataset = imu_collector.collect_many_classes(
        dataset_name='ContinuousMotion', 
        duration=30
    )
Activeu l'entorn 'ml' (machine learning / aprenentatge automàtic) del conda:
 conda activate ml
Us ha de sortir '''(ml)''' a l'esquerra del terminal. Executeu el programa '''collector02.py''' (amb la modificació del nom del port de comunicacions al vostre sistema operatiu. Per defecte hi ha '''/dev/ttyUSB0'''). Per a fer la captació de les dades he anomenat quatre estats gestuals ('''quiet''': placa quasi quieta. '''amunt-avall''': movent la placa amunt i avall. '''esquerra-dreta''': movent la placa a esquerra i dreta. '''cercle''': movent en cercle la placa):
 (ml) $ '''python collector02.py''' 
 This is an interactive data capturing procedure.
 Keep in mind that as soon as you will enter a class name, the capturing will start, so be ready!
 Which class are you going to capture? (leave empty to exit) '''quiet'''
 31it [00:30,  1.01it/s]                                                                                 
 Captured 1805 samples
 Is this class ok? (y|n) y
 Which class are you going to capture? (leave empty to exit) '''amunt-avall'''
 31it [00:30,  1.03it/s]                                                                                 
 Captured 1771 samples
 Is this class ok? (y|n) y
 Which class are you going to capture? (leave empty to exit) '''esquerra-dreta'''
 31it [00:30,  1.03it/s]                                                                                 
 Captured 1777 samples
 Is this class ok? (y|n) y
 Which class are you going to capture? (leave empty to exit) '''cercle'''
 31it [00:30,  1.02it/s]                                                                                 
 Captured 1794 samples
 Is this class ok? (y|n) y
 Which class are you going to capture? (leave empty to exit) 
 Are you sure you want to exit? (y|n) '''y'''

i genera l'arxiu '''imu.csv''' (exemple d'arxiu [https://github.com/jordibinefa/IoT-02/blob/master/codes/python/machineLearning/imu.csv imu.csv] generat).

Principi de l'arxiu '''imu.csv''':
 (ml) $ '''head imu.csv''' 
 ax,ay,az,gx,gy,gz,target,target_name
 0.1,0.08,9.71,-0.01,-0.05,-0.07,0.0,quiet
 0.14,-0.07,9.64,-0.0,-0.06,-0.06,0.0,quiet
 0.12,-0.1,9.6,-0.02,-0.05,-0.06,0.0,quiet
 0.11,-0.04,9.65,-0.02,-0.05,-0.06,0.0,quiet
 0.09,0.0,9.7,-0.01,-0.05,-0.06,0.0,quiet
 0.12,-0.02,9.66,-0.02,-0.05,-0.05,0.0,quiet 
 0.14,-0.02,9.69,-0.01,-0.06,-0.06,0.0,quiet
 0.1,-0.02,9.64,-0.01,-0.06,-0.06,0.0,quiet
 0.09,-0.04,9.64,-0.02,-0.05,-0.05,0.0,quiet
Final de l'arxiu '''imu.csv''':
 (ml) $ '''tail imu.csv'''
 0.05,1.05,9.84,0.02,-0.24,0.66,3.0,cercle
 0.01,1.3,10.63,0.08,-0.21,0.7,3.0,cercle
 -0.06,1.5,10.86,0.05,0.03,0.67,3.0,cercle
 -0.08,1.65,10.44,0.06,0.29,0.61,3.0,cercle
 0.05,1.58,9.56,-0.02,0.45,0.56,3.0,cercle
 0.1,0.84,8.74,-0.07,0.36,0.46,3.0,cercle
 0.49,0.77,8.32,-0.01,0.05,0.53,3.0,cercle
 -0.1,0.92,9.93,0.09,-0.26,0.43,3.0,cercle
 -0.03,1.17,10.75,0.26,-0.29,0.46,3.0,cercle
 0.13,1.34,10.8,0.27,-0.09,0.38,3.0,cercle
Si l'arxiu '''imu.csv''' ja existeix es visualitza el resum de les dades enregistrades al tornar a executar '''collector02.py''':
 (ml) $ python '''collector02.py''' 
                 ax           ay           az           gx           gy           gz       target
 count  7147.000000  7147.000000  7147.000000  7147.000000  7147.000000  7147.000000  7147.000000
 mean      0.126964    -0.001574     9.631492    -0.016372    -0.059793    -0.079418     1.498111
 std       1.517042     0.885516     1.053995     0.248000     0.157644     0.412540     1.121298
 min      -7.180000    -2.770000     4.330000    -1.230000    -0.990000    -1.490000     0.000000
 25%      -0.590000    -0.640000     9.480000    -0.100000    -0.110000    -0.250000     0.000000
 50%       0.140000     0.040000     9.650000    -0.020000    -0.050000    -0.120000     1.000000
 75%       0.480000     0.490000     9.810000     0.060000    -0.010000     0.120000     3.000000
 max       6.070000     3.390000    14.180000     1.190000     0.690000     1.240000     3.000000

= Generació de codi per l'Arduino IDE =

A l'executar [https://github.com/jordibinefa/IoT-02/blob/master/codes/python/machineLearning/imu2arduinoIde.py imu2arduinoIde.py] (IMU to Arduino IDE) es generen dos arxius: '''Classifier.h''' i '''Pipeline.h''' ([https://github.com/jordibinefa/IoT-02/tree/master/codes/python/machineLearning/newArduinoIdeFiles exemples de Classifier.h i Pipeline.h] a partir de l'exemple [https://github.com/jordibinefa/IoT-02/blob/master/codes/python/machineLearning/imu.csv imu.csv])

Aquests dos arxius es generen a una nova carpeta anomenada ''newArduinoIdeFiles'' que hauran de ser copiats al mateix directori que l'arxiu .ino

Arxiu [https://github.com/jordibinefa/IoT-02/tree/master/codes/IoT-02_IMU_Classify_by_text IoT-02_IMU_Classify_by_text.ino] emprant Classifier.h i Pipeline.h generats amb l'arxiu imu.csv

Arxiu [https://github.com/jordibinefa/IoT-02/tree/master/codes/IoT-02_IMU_Classify_by_led IoT-02_IMU_Classify_by_led.ino] que detecta el gest i l'identifica amb un led de color.
<!-- = Instal·lació de l''''edge-impulse-data-forwarder''' =
Aquesta eina serveix per a publicar dades des de la placa fins al servidor d'[https://edgeimpulse.com/ Edge Impulse]

 curl -sL https://deb.nodesource.com/setup_16.x | sudo -E bash -
 sudo apt-get install -y nodejs
 mkdir ~/.npm-global
 npm config set prefix '~/.npm-global'
 echo 'export PATH=~/.npm-global/bin:$PATH' >> ~/.bashrc
 npm install -g edge-impulse-cli --force
Un cop instal·lat, es pot executar des del terminal:
 edge-impulse-data-forwarder
-->

= Bibliografia =
[https://eloquentarduino.com/gesture-classification/ Gesture Classification] by [https://eloquentarduino.com Eloquent Arduino]

[https://www.hackster.io/Yukio/gesture-classification-with-esp32-and-tinyml-dab252 Gesture Classification with Esp32 and TinyML] by [https://www.hackster.io/Yukio João Vitor Yukio Bordin Yamashita]