View Past Scans and Predictions
LiDAR / Photo Input
View Prediction
Labelled Points of Interest with Custom Trained YOLO v8 pose estimation model
Labelled Points of Interest with Custom Trained YOLO v8 pose estimation model
Additional feature input
Example of LIDAR data
Example 2 of LIDAR Data
Example 2,Labelled Points of Interest with Custom Trained YOLO v8 pose estimation model

Cowdar

Inspiration

Our research into cattle management revealed that existing automated weight prediction systems for livestock often fall into one or more of these categories:

Prohibitively expensive
Subscription based
Low quality

Many government-funded projects in this space have been abandoned or remain unavailable to the public. We wanted to bridge this gap for small time farmers by creating a free, open-source tool that leverages the hardware already in their pockets.

What it does

Estimates the weight of cows using a LiDAR snapshot and a photo.

How we built it

Trained a custom yolov8m-pose model with 100 epochs using a dataset of labelled cow poses created by Sorin Workspace — 1042 total images (729 training, 209 validation, 104 testing) — to analyse points of interest on the cow (approx 30 min on Nvidia T4)
- This model outputs a set of normalised coordinates (normalised refers to a relative x,y value range of 0–1, as opposed to pixel measurements)
The 3D LiDAR point cloud is mathematically projected onto a 2D image plane
- The server matches the normalised coordinates to the closest projected 3D LiDAR points to determine (estimate) the physical distance between points of interest on the cow
- 3D Euclidean distances are extracted between these coordinates (e.g., body length and radius)
- Weight is predicted using a variant of Schaeffer's Formula

Challenges we ran into

Figuring out how to get LiDAR data of a moving object
The dark hide of Angus cows is much harder to parse points of interest on
Our limited knowledge with maths and AI limited the approaches we could use to predict weight
Labelled cow pose datasets were limited in availability, quality, and size — the time scope of our project meant we couldn't hand-label poses on Angus cows
Schaeffer's Formula doesn't account for features like body fat, breed, sex, or age

Accomplishments we're proud of

Successfully creating a functional proof of concept that integrates LiDAR hardware sensors with modern AI models
Developing a pipeline that transforms raw LiDAR and photo data into a meaningful physical measurement (weight)
Keeping the project entirely open source to ensure it remains accessible to the farmers who need it most
Developing an extensible project