AI Image Rating for Construction Timelapse: How Our Optimized AI Filters Out Fog, Rain and Poor Visibility

Anyone who has run a construction camera for a few months knows the problem. Between the clear days there are foggy mornings, raindrops on the lens and days where you can barely see the site. If those images end up in the final video, it starts to flicker and loses its flow. That's why TimelapseRobot automatically rates every image with 1 to 4 stars based on how well the subject is visible. We've now reworked this classification with our own optimized AI classification system, and it rates images more accurately and reliably than before.

How the star rating works#

Every image gets a rating between 1 and 4 stars. 1 star means poor visibility, for example dense fog or drops on the lens. 4 stars means the best possible view of the site.

The rating isn't about composition or aesthetics. It only answers one question: how well can you see the subject?

Less flicker, cleaner videos#

The star rating becomes useful when you render a timelapse video or create a ZIP download. In the Gallery, you can filter your images by star rating, so images with poor visibility are left out automatically.

The result is a video without sudden jumps between clear and hazy frames. Less flicker, smoother footage, and no hours spent sorting out bad images by hand.

What changed with the new system#

The star classification itself isn't new. What's new is the system behind it. It builds on an established, openly available vision AI model, which we optimized specifically for construction timelapse images and validated against more than 12,000 hand-rated images.

The new system is already live. Since September, all images from Business accounts are rated with it automatically.

It classifies around 86% of images correctly. That's a clear step up in reliability compared to the previous version.

Built on more than 10 years of construction timelapse#

We've been producing construction timelapses ourselves for more than 10 years. Over that time we've collected a large number of real datasets from construction sites in all kinds of weather and lighting conditions.

That's what made this optimization possible. We use this data to tune and validate the classification on exactly the situations that matter on a real site: fog, rain, low sun, dirty lenses. A generic vision model doesn't know what a construction site looks like on a grey November morning. Ours has been optimized on thousands of them.

What's next#

We're not done. Our goal is to keep pushing the accuracy further. We'll continue working on the system and release improved versions over time.

Try it yourself#

The star rating is available in the TimelapseRobot Business plan. If you're on Business, your images are already being rated by the new system. Just open the Gallery, set the star filter and render your next video or ZIP download. And if you're not on Business yet, this might be a good reason to take a closer look.