parkoff 0.1.0

No project description provided

Problem Statement:

Park Off

Finding a parking spot in downtown of any city is a major issue and especially with Return to Office policies, this can even intensified. Moreover, if you live in downtown and prefer to walk, unruly weather is quite an annoyance. Here is an attempt to use Traffic Cams to find parking spots and street wise weather updates in near real time.

Introduction:

Seattle Gov has an extensive array of Traffic Cameras spread across the city that update the image and sometimes videos every minute.

The cameras are located on all major jucntions and crossings and have a decent picture as you can see

This project aims at querying this data, make a cache, and then run some object detection model on these images to find cars parked by the curbs.

A stretch goal is to detect the weather conditions by these images and get a near real time update if it is convinient to walk down a few blocks.

Finally, this could all be overlaid as a custom layer on Open Street Maps.

Project Outline and Goals:

• Pipeline deployed on Raspberry Pi that downloads these Traffic Images near real-time and caches locally ☑
• Vehicle detection using YOLO NCNN model ☑
• Curb Detection using Canny Edge Detection ☑
• Finding Parked Vehicles ☑
• Spot Empty Spaces for Potential Parking ☐
• Identify Rainy or Not Rainy Weather ☐
• Overlay Traffic Cam Position with the GIS Parking Data ☐
• Open Source Python Package for others to join in ☐
• App and Backend ☐

Constrains:

• Every Traffic Cam updates 1 min
• Glare often spoils the picture quality
• Perspective of Traffic Cams oftens makes it hard to identify Gaps in between cars

Work Breakdown

Feature	Status
YOLOv11 Test	☑ Completed
ChatGPT Test	☑ Completed
Seattle gov image scrape	☑ Completed
Automated Scraping	☑ Completed
JSON Project Configuration	☑ Completed
Filter Yolo Objects	☑ Completed
Fine Tune Yolo detector	☑ Completed
Test ChatGPT for the above	☑ Completed
Canny Edge Detection for Curbs	☑ Completed
Hough Lines to Find Long Stretches of Curbs	☑ Completed
Fine Tuning Curb Detection	☑ Completed
Filter Curbs Based Upon Color	☑ Completed
Find Vehicles Nearest to Curbs	☑ Completed
Subset Parked Vehicles	☑ Completed
Find Gap between/Along Curbs	☐ In progress
Weather Detection	☐ To do
Cache Parking And Weather for each cam	☐ To do
Overlay on OSM	☐ To do

Design Flow

Folder Structure

parkoff/parkoff: Contains Functions and definition

parkoff/tests: Contains Unit Test Case

parkoff/utils: Contains Service Script for Raspberry Pi Deployment

Installation

Install Python Dependency Manager

Refer : https://python-poetry.org/docs/

Install Local Dependencies

This should create a local env variable and install all the libs

cd parkoff
poetry install

Build Package

This should create *tar or *whl for us to invoke in any application under dist directory

poetry build

Test Package

Local Test of the package

poetry run pytest

Usage

PC Development

Get the env variable name

poetry env info

• Use this variable for your py-kernel when you launch your jupyter instance in vscode.
• If you cant find this, then:
• Ctrl/Cmd + P
• Select Python Interpreter
• Enter Interpreter Path
• Use the path as per poetry env info to add your venv to vs code

RPI Deployment

Setting Up the PI

• Install 64-bit OS (Bookworm) in headless version
• On RPI after successful SSH, create two folders to abstract code and data for cleanliness and easier management

mkdir -p Work
mkdir -p Data

Next we will copy the code base into RPI using rsync

rsync -a --exclude="/.*"  $GitClones/parkoff/parkoff/parkoff  pi@192.168.4.49:/home/pi/Work/parkoff/parkoff/

This ensure we only copy the code and nothing else

A quick test can be performed by running a sample test script before we automate the pipeline But before that, we need to create a virtual environment

cd Work/parkoff
python -m venv .parkoff
source .parkoff/bin/activate

We need the packages that parkoff requires and that can be installed on RPi in two ways

• Using poetry on RPI ( requires installation and maybe heavy)
• Export the package information as a 'requirements.txt' and then install in the env variable

The latter is easier and more lightweight.

To export the packages on your host, navigate to the pyproject.toml and then export

cd  parkoff/parkoff
poetry export --without-hashes > requirements.txt

You will now see a requirements.txt with all the packages. Let's resysnc the repo with the RPI.

rsync -a --exclude="/.*"  $GitClones/parkoff/parkoff/parkoff  pi@192.168.4.49:/home/pi/Work/parkoff/parkoff/

Now let's install the packages

cd Work/parkoff
python -m venv .parkoff
source .parkoff/bin/activate
pip install -r parkoff/requirements.txt

If all goes well, you are ready to test the code.

python parkoff/tests/test.py

Once run, it should cache some output.

Pipeline Setup

Python Scheduler is a OS agnostic framework. It helps automate scripts periodically.

parkoff/parkoff/scheduler.py: calls the relevant functions every 60s to cache and analyse the data.

Next Linux Native System Service was used to run this scheduler as the RPI reboots.

A sample service file is provided here utils/parkoff.service

[Unit]
Description=Parkoff Scheduler Script
After=network.target

[Service]
Type=simple
ExecStart=/bin/bash -c ' source /home/pi/Work/parkoff/.parkoff/bin/activate  &&  python scheduler.py --logpath /home/pi/Work/parkoff/. >> /home/pi/Data/parkoff/logfile.log 2>&1'
WorkingDirectory=/home/pi/Work/parkoff/parkoff/parkoff
User=pi
Group=pi
Restart=always
RestartSec=5

[Install]
WantedBy=multi-user.target

To create a service file, run the following and copy paste the above code into it

sudo nano /etc/systemd/system/parkoff.service

Next, we will enable the service and turn it on

sudo systemctl daemon-reload
sudo systemctl enable parkoff.service
sudo systemctl start parkoff.service

You can view the status of the service

sudo systemctl status parkoff.service

Or even stop it

sudo systemctl stop parkoff.service

Next we will modify some of the file paths in the project settings.json to fit the rpi

{
    "settings": {
        "model":{
            "name":"yolo11n.pt",
            "format":"" ,
            "confidence" : 0.2,
            "iou": 0.5
        },
        "curb_detection":{
            "blur_box_edge" : 3,
            "canny_thres_0" : 30,
            "canny_thres_1" : 150,
            "dist_bw_curb_car" : 30,
            "color_gradient_thrs":200,
            "min_straight_line": 20,
            "max_gap_bw_lines":20

        }
    },
    "img_cache":{
        "input_dir": "/home/pi/Data/parkoff/input/",
        "output_dir": "/home/pi/Data/parkoff/output/"

    }

Once saved, we can restart the parkoff.service

sudo systemctl restart parkoff.service

Once done, you should be able to see the raw images being cached in "/home/pi/Data/parkoff/input/" and the output being cached at /home/pi/Data/parkoff/output/ with all the logs at /home/pi/Data/parkoff/

Monitoring and Health

Circular Logs

The scheduler defines a circular logger to log all input of status

This saves at max 3 files of 1MB each.

You can monitor this in runtime by running

tail -f Data/parkoff/app_parkoff.log

Temperature Monitoring

There is built in temperature monitoring. Before every anaylysis of any image, the temperature is checked and only if the temperature is <75 degC the images are analysed. The below logs show how the temperature increases as every image is analysed.

Here is a look at the logs on RPi

In case the temperature is over the set threshold, the service skips analysing the images and only caches the files. It is designed to come back for analysing at a later point in time.

Service Monitoring

You can monitor the status of the service and it's logs by running

sudo systemctl status parkoff.service

tail -f Data/parkoff/logfile.log

...

Contributing

Mass Production Readiness:

Conclusion:

References:

• Wikipedia
• RaspberryPi
• StackOverFlow