Inventory Management using Video Analytics

Building a complete inventory management solution generally involves connecting to an ERP or inventory management system. However, a good first step in building such a system is to train visual detectors to identify objects of interest, and annotate a video so you can see how well it's working. This demo will walk you through the process of training a detector and annotating a video.

Click here for full resolution

Installation

The code supports installing the dependencies using uv.

# Clone this repo
git clone [email protected]:groundlight/inventory-management.git

Then install the dependencies using uv

cd inventory-management
uv venv
uv sync --no-build-isolation

Generating the data

Videos

Download them, and put them into data/videos and DVC.

Frames

First, consider how much motion detection you want, and preview. Something like 0.1% will be very sensitive, and catch any tiny motion. Something like 2% will catch fewer more representative frames.

./video-to-frames.py data/videos/videoname.mp4 \
    --pct-threshold 0.1 \
    --preview

Run the script to generate the frames:

./video-to-frames.py data/videos/videoname.mp4 \
    --pct-threshold 0.1 \
    --save-to data/frames/videoname-0.1

The generated frames will be saved to data/frames/videoname-0.1