Making an long-term outdoor timelapse video is challenging due to many uncontrolled phenomena (lighting, weather, raindrops and dirt, irregular content or shooting frequency, etc.). A perfect timelapse should be able to provide a maximum of information in a pleasing and consise video. To this end, we develop different strategies seeking to reduce non-informative distractions without destroying informative content.


method information distraction
smart selection
stabilization -
relighting -
enhancement -


Smart Selection

Most of the time, the generated video is short enough to require fewer images than we have. Yet, not all photos are equal. We have developed a custom optimization algorithm able to select the optimal image sequence that best satisfies a set of constraints. The selection takes advantage of the photo metadata as well as many attributes measured by our AI engine like raindrops, weather, activity. It also seeks to minimize the flickering by imposing a low lighting distance between consecutive images.



Smoothing is the blending together of neighboring frames. This is a simple yet effective technique for reducing flickering. However, this is at the cost of a loss of information and increased redundancy between frames. It also has the disadvantage of generating semi-transparent objets called ghosts. Our advice is to use it as the last resort for killing remaining flickering.



Many fators can lead to an unstable video (wind, vibrations, thermal expansion, progressive collapse, etc.) and therefore, it is common to have shaking despite precautions. However, stabilizing timelapse videos is more challenging than standard videos due to substantial visual change between consecutive frames. For this reason, usual technics do not work and we developed a custom stabilization algorithm featuring 3 different lens model to cover Tikee's wide angle as well as other cameras.



When frame selection is not sufficient to reduce flickering, typically for mid-term timelapses, there is no choice but to color-align the frames. This is a key feature to deliver pleasing timelapses and conventional anti-flicker technics are not sufficient as they only apply global corrections. We are constantly improving our original AI-based approach which is not only able to align global illuminations but also apply local corrections allowing significant attenuation of projected shadows. The model particularly targets recurrent phenomena and reduces them while preserving ephemeral ones, often carrying meaningful information.



Unlike a photographer, your timelapse camera shoots images with the same framing in all the conditions. Thus, some photos are inevitably taken in non-optimal lighting conditions. We leverage a set of image processing algorithms to make all the areas well visible whatever their illumination.


Video credits: ELOY Belgique