Microsoft's 'hyper-lapse' videos create buttery smooth 3D worlds from your photos

The software is designed to take jittery footage from wearable cameras like the GoPro (above) and turn it into something beautiful (below)

Click to follow
The Independent Tech

Microsoft has shown off a new set of algorithms that can turn any jittery sequence of photos into an eerily smooth time-lapse video.

Developed by the company’s research labs, the technique processes images snapped by head-mounted cameras to create a detailed 3D map of the world for a computer to walk through.

Each frame is analysed individually to create this virtual world, with the algorithms then creating out a path of best fit that follows the photographer's route as smoothly as possible.

The results are entrancing but slightly unreal - a side effect of the blending techniques means that textures in hyper-lapse videos tend to pop in and out of focus; a visual phenomenon familiar to gamers, for whom limited processing power can mean a world view that loads section by section.

The team behind the software say they want the algorithms to take advantage of footage shot by wearable cameras like the GoPro and Google Glass, turning mundane time-lapses (such as cycling through a city) into something infinitely more watchable.

These sorts of hyper-lapse videos might become common in the future then, but there’s still one big hurdle to overcome: turning a ten minute video into a smooth hyper-lapse took the researchers 300 hours of computing time - although they got this down to simply ‘hours’ after the algorithms were finessed. Still, don't expect to see many hyper-lapse gifs for a couple of years yet.