Bin Fill Level Estimation

Project overview

A dataset preparation project supported an AI-based bin fill level estimation system. The task involved reviewing waste bin images manually, estimating the percentage of bin occupancy, and updating each image filename with the corresponding fill level.

Each image was inspected visually and the estimated fill percentage was appended to its filename following the client's naming convention, ranging from empty bins at nought per cent through to overflowing bins above one hundred per cent. The prepared dataset was used for machine learning model development.

Work performed

• Reviewed the complete image dataset manually.
• Estimated the fill level of each waste bin from its visible contents.
• Renamed image files by appending the estimated fill percentage.
• Consistent estimation was maintained across the entire dataset.
• Filenames were verified for accuracy before final delivery.

What made it hard

• Keeping fill level estimation consistent across thousands of images.
• Camera angles, lighting conditions and bin types varied widely.
• Partially filled bins had to be estimated with high accuracy.
• Filename consistency had to match the client's naming convention exactly.

Result

A high-quality fill level estimation dataset was prepared by assigning fill percentages manually to all images. The renamed dataset was delivered to the client's specifications and was ready for AI model training.

Final deliverable

Manual fill level estimation was completed for the entire dataset, with the occupancy of each waste bin estimated accurately and the corresponding percentage added to the image filename. The final dataset was organised, consistent and ready for machine learning applications.