Bikeworkz

Installation

pip install -r requirements.txt

Demo

cd src
python demo/Streaming_4K_Camera.py
python demo/Object_Detection_RGB.py
python demo/Object_Detection_Left_Right.py
python demo/Monocular-Neural-Inference-fused-with-Stereo-Depth.py

First attempt to run on raspberry pi: apple.py

A simple script that will recognise a bottle (the pretrained model does not recognise apples it turns out). The output of this script is whether or not a bottle is on screen, the location of the bottle (left, center and right) and the distance. These metrics should be written to a file where another running process can pick up this output and control the output pins of the Raspberry Pi that are connected to LED’s. The end product should be a device that will light up when a bottle is in view. If it is in the middle the signal will change/get more intense.

How to install and use

On the Raspberry Pi, install a recent version of python (3.9 in our case). And use the following to install the dependencies for depthai, the python package that controls and reads from the camera:

sudo curl -fL http://docs.luxonis.com/_static/install_dependencies.sh | bash

Once the dependencies are installed, run:

pip install depthai

In a (virtual) python environment. If depthai is successfully installed the script to start the camera and start detecting the bottles:

python src/apple.py

This should start writing detections (of bottles) to a file.

Notes

To run examples in depthai-python, first run:

git clone [email protected]:luxonis/depthai-python.git
cd depthai-python/examples
python install_requirements.txt

• Make sure Raspian OS is up to date (libc6)
• Python 3.9
• libc6: >= 2.30
• glibcxx_3.4.26

2021-10-16, 16:42 - Something is working

The script src/pears.py is now able to run on a Raspberry Pi. This means that we can run object detection on the camera and periodically update a file that will either contain a simple 0 or a 1. The script src/joep.py will continually read the output file from src/pears.py and based on the number turn on an orange or a red LED.

2021-10-16, 21:32 - Nothing is actually working

We decided to take the next few steps: premature optimization and feature bloat. As was expected: this did not lead to a better product. In hindsight we should have simply milked the working prototype and left it there. But no, beer, bacardi cola made us over confident. We wanted exotic LED blinking patterns, depth and speed detection of our objects... on a Raspberry Pi. Does this sound like a project suited for 3 increasingly intoxicated programmers? No. Like Icarus we flew too close to the sun, we were too ambitious, we've failed....

Communication

Communication between camera (script) and arduino is done by writing to a file. This file contains two digits, which will determine the combination of LED's turning on:

• Position 1:
  • 0: No object detected
  • 1: Object detected on the left
  • 2: Object detected in the center
  • 3: Object detected on the right
• Position 2:
  • 0: Moves away from me - warning - green/yellow
  • 1: Moves closer to me (slowly) - yellow/red
  • 2: Moves alarmingly fast closer to me - red/blinking