pedestrian-detection

Computer Vision Project using YOLOv8 to detect and count the number of people in the livestream from Illinois Quad Cam and Alma Mater

Setting up Environment

Windows (WSL2)

1. Launch your instance of WSL2
2. Install Mamba and depending on whether you have Anaconda installed or not, there are two different sets of instructions. We are using mamba as it's much faster than conda at installing packages.
3. git clone https://github.com/yaswant2403/pedestrian-detection
4. Then open your bash terminal in the folder you cloned your repo in and run the following set of commands one by one

mamba env create -f environment.yml # creates the environment according to the yml file
mamba init

Close your terminal and reopen it.

mamba activate pedestrian-detection # activates environment so that the packages are usable in the code

5. In backend/ run extract_images.py to download the video and extract the images into data/video/frames/. You'll need to change some filepaths. Refer to Configuration

   For development environment, don't set the duration more than 120 seconds.

6. Split your terminal (In VSCode) or open a new terminal and run run_inference.py to output the number of detections that standard yolov8n.pt runs. In the future, we will have our own model trained on a custom dataset.

Mac

1. Install Mamba and depending on whether you have Anaconda installed or not, there are two different sets of instructions. We are using mamba as it's much faster than conda at installing packages.
2. git clone https://github.com/yaswant2403/pedestrian-detection
3. Then open a bash terminal (Mac may sometimes open zsh which we DONT want) in the folder you cloned your repo in and run the following set of commands one by one

mamba env create -f environment.yml # creates the environment according to the yml file
mamba init

Close your terminal and reopen it.

mamba activate pedestrian-detection # activates environment so that the packages are usable in the code

4-6) Same steps as Windows

Configuration

You can use whereis ffmpeg and whereis yt-dlp to get your specific filepaths. Then, replace all the /home/yash/miniconda filepaths in extract_images.py with your specific fielpaths of ffmpeg and yt-dlp.

If you want to download the video file for more than 10 seconds, set the duration variable number to a different value.

Additionally, if you only want to extract images every 5/10/15 seconds instead, change the fps=1 argument in extract_images.py to fps=5,fps=10,etc

Running Project

First, git clone https://github.com/yaswant2403/pedestrian-detection.git

Currently, you will have to run two different files. The first file is extract_images.py which will download the livestream of the Quad Cam to video/raw-mp4 for x seconds and then extract the frames into video/frames. The second file is run.py which will launch the website and run inference on the images in the frames/ folder displaying that in the heading.