Special Topic of Tzu-Chi at TCIVS

Purpose of this file

• I just want to write up some problems while I set up the environment and hardware of this project

Hardware info.

Spec.	Raspberry Pi 3 Model B+
CPU	ARM Cortex-A53 1.4GHz
RAM	1GB SRAM
Wi-Fi	2.4GHz and 5GHz
Ethernet speed	300Mbps
Bluetooth	4..2

Set up sequence(Ideal)

• Install OS to Raspberry pi

  You can just check this page

   $  vim /etc/netplan/50-cloud-init.yaml
   (add the line at the end, and the indentation is very important)
   wifis:
   	wlan0:
   	dhcp4: true
   	optional: true
   	access-points:
   		"home network":
   			password: "123456789"
   $  sudo reboot
  

• Install Anaconda in a correct version

   $  cd ~
   $  curl -O https://repo.anaconda.com/archive/Anaconda3-2021.04-Linux-aarch64.sh
   $  bash Anaconda3-2021.04-Linux-aarch64.sh
   $  vim ~/.bashrc
   (add "export PATH='/home/ubuntu/anaconda3/bin:$PATH' at the end")
   $  source ~/.bashrc
   $  sudo reboot
  

• Install the Library you need

   $  conda install -c anaconda scipy
   $  conda install -c conda-forge/label/broken tensorflow
   $  pip install opencv-contrib-python
   $  conda install -c anaconda numpy
   $  conda install -c anaconda requests
   $  conda install -c conda-forge keras
   
   $  conda install -c conda-forge imutils
   $  conda install -c conda-forge face_recognition
   $  conda install -c conda-forge dlib
  

• Run the python file you mount from the external disk, e.g. flash disk

Problem

1. There're 3 different OSs can choose including Raspberry Pi OS, Ubuntu server, Ubuntu desktop

   • First of all is Raspberry Pi OS(32-bits), because it's an official recommendation, I install it first. But as I show the spec. above, the CPU is 64-bits and you must run Raspberry Pi Imager before you install OS to Raspberry Pi. Then your OS architecture will be not compatible with Anaconda though it has a 32-bits version as well. You'll get the error.
   • The second one is Ubuntu-Desktop (22.04 or 20.04). It'll get frozen all the time because of the small SRAM with 1GB
   • The third one seems quite a good choice as an OS. It'll not get frozen because it's just a simple CLI system and it also has an aarch64 version. Then the statements below are the problems you'll encounter.

2. If I install Anaconda correctly, I'll encounter a problem that conda create instruction can not be used. You'll get an error message like this: Illegal instruction(core dumped).

   • Solution by installing Miniconda.

   • But...miniconda still has another problem: version conflict with the library. So, this is not the best solution as well.

   • For more information on this solution: though I can use conda create instruction, I can not install python with the 3.6 version. The reason that I must install this version is for the library I want to install later. If I don't install version 3.6, I can't install imutils, face_recognition, and dlib at the same time. The other library list above including scipy, TensorFlow, NumPy, and so on will install successfully in versions 3.6 to 3.9.

   • Briefly speaking, because of my OS architecture, I can't install these 3 libraries by the statement that anaconda official supplied. I can install the package available on noarch or aarch64 platform only.

   • For imutils, like the image below(img1) $ conda install --channel https://conda.anaconda.org/gilbertfrancoins imutils

   • For face_recognition, like the image below(img2) $ conda install --channel https://conda.anaconda.org/conda-forge face_recognition

   • You can check the error on this page, then there is another problem I encounter is I can not use anaconda instruction to search the library package. So, I use my laptop(a normal win10 system) to search.
   • BTW, you can not use the x86 version, because it'll crash while the installation

---

---

---

3. I also followed this article and tried to address this problem. Though it can use anaconda instruction smoothly, it still has some problems to solve(I forgot the problem, QAQ)

4. You might be wondering why I don't use pip instruction. Because you'll get an error message like this: Illegal instruction(core dumped).

5. Other problems must address

   • If you install OS and Anaconda successfully.

   $  python
   $  import numpy
   
   (error message)
   $  Illegal instruction(core dumped)
   

Conclusion for the above

The solutions above are not suitable for this project

New Solution

This solution seems fine so far, so I write it up as below

1. First, we can install Raspberry Pi OS (64-bit) by Raspberry Pi Imager. It has a desktop version and is still compatible with the hardware.
2. Second, install Miniconda by following the instruction on this page(PS version is Miniconda3-py37_4.9.2-Linux-aarch64.sh)
3. Third, create a new environment in Anaconda without python. You should install python independently(v3.6).

   $ conda install -c moussi python
   $ conda install -c akode face_recognition_models
   $ conda install -c gilbertfrancois imutils
   $ conda install -c conda-forge fortran_stdlib
   $ conda install -c jetson-tx2 scipy
   $ conda install -c intel tensorflow-base
   $ conda install -c anaconda numpy
   $ conda install -c conda-forge/label/cf202003 requests
   $ conda install -c conda-forge keras
   $ pip install opencv-contrib-python
   

   These libraries can be installed with python=3.6, but TensorFlow. Please go to this page and search the library you want to install(set the platform filter as noarch or Linux-aarch64)

Practical Solution

In order to avoid not being able to do it in the end, we change another solution with higher success rate - we used Arduino instead. You can check the code in here. And our os platform is my x86 laptop, we don't have the software compatible problem.

Reference

使用Python的pySerial模組進行序列通訊