Classificació de gestos emprant la placa IoT-02
Contingut
Instal·lació d'eines d'aprenentatge automàtic (machine learning) i tensorflow emprant Python
Si no teniu instal·lat Anaconda, feu-hi 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 IoT-02_mpu6050_dataForwarder.ino
Canvieu el port de comunicacions (a Linux /dev/ttyUSB0) d'aquest tros del codi 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):
(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 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
Instal·lació de l'edge-impulse-data-forwarder
Aquesta eina serveix per a publicar dades des de la placa fins al servidor d'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
Gesture Classification by Eloquent Arduino
Gesture Classification with Esp32 and TinyML by João Vitor Yukio Bordin Yamashita