MotionDectect

~~ We are using Kanban development: Situations of developing shoud be updated at the top of logs ~~

** We decide to use Iteration design pattern, when updated, a log record must be added as a reminder to each other **

Status:

Project Stage: Operational Protrotype (installation of torch in Pi failed. Test in laptop) visit https://riino.site/2019/12/27/How-to-set-up-current-work-of-MotionDetect-Dev.27th/#how-to-run to check. **Final Report is delay. New DDL : 1.14 16:00

Team number:

• Javier Sanguino：

  • fixed work time
    • Monday: 12:00-18:00
    • Friday: 15:00-20:00
    • Saturday: 10:00-14:00
    • Sundays: 10:00-14:00
  • final day for flight:
    • 10th, Jan. 2020.

• Riino(aka Shih-Chi CHANG):

  • fixed work time:
    • Thur.: 13:00--17:00
    • Fri.：13:00--17:00
    • Weekends: Anytime

Currently goal:

1. get info from the sensor
2. send data into Pi.

Logs:

• log 12.17 :

  Inited github readme. We are going to set a Kanban(a real board in the lab) to show the process and schedule.

• log 12.19 : Going to establish a simple model to transfer data from STM to Pi using socket,which should includes a .py file in each side called link.py

• log 12.20 The remote repo of Pi is located on home/git/

• log 12.23 Data from the STM32 is accessed continuously. Problems debugguging, ready to merge with sockets for communication with RaspPi

• log 12.27 （mid-term report) All basic functions are Successfully done: WIFI,socket,censor of STM32, WIFI,socket on Python. I use my laptop to edit/run python. Those codes should be easy to just copy and run in Pi.

• log 1.9 We made some extra format adjustment to keep the length of data that everytime STM32 sends fixed. On 1.9, we have store_recieve.py to pick a fixed length of data stream from STM32,and save the data as a .csv file with a input lable PASS : 0, SHOOT: 1 for training. Also, the installation of pytorch in Pi is running. According to doc it will take 3 hours.

• log 1.10 (final report in schedule) We used about 2 and a half hours , to collect 250 sets of data aof gyro sensor. Each one with 49 records in 3 seconds. Still installing pytorch in RasPi.

Javier flew back. Distance working now.

• log 1.12 Installing PyTorch in Pi, which needs e a whole day and night. Also, Model trained over 225 sets of data we took on 1.10. Model tested over 25 random picks of sets of data. Model accuraccy: 70%. We also prepared Test.py prepared for predicting the kind of movement on the Pi.

• log 1.13 After over 72 hours of Installation of torch in Pi, we moved platform to laptop for testing. After some configuration, our ML models works and give its output between 'pass' and 'shoot'.

• log 1.14 (final report) Optrational classifier done. TODO: Design a threshold to make auto-detection.

Demo

Pi side : Run MotionDetect.py (need models.py and other module ) STM side : Run main.cpp

You need to change IP address in both Pi and STM side before running. To install torch in Pi Run night.sh in Pi. e.g. sudo bash night.sh