Lightly reduces data redundancy for autonomous driving

(CONNECT) Lightly wants to support European car manufacturers in the race for autonomous driving with its new data collection solution LightlyEdge. According to a press release, LightlyEdge uses so-called edge AI directly on the cameras and sensors of vehicles to enable data capture and video analysis by artificial intelligence (AI) at the point of capture in real time and without cloud infrastructure. It automatically identifies valuable, critical scenarios, while redundant data is ignored. By using LightlyEdge, companies could "stay ahead of the global competition by implementing real-world scenarios faster" and save costs and data storage at the same time.

This is Lightly's response to the fact that "the autonomous driving and edge AI industry is at a turning point": although vehicles and devices are more powerful than ever, the data pipelines have not kept pace. "Developers are still forced to either log everything (and get lost in data) or rely on simplistic triggers that miss important events."

In contrast, LightlyEdge can use diversity selection to determine so-called frames that are visually different from those already saved, such as a new construction site on the road. As the company, which was founded in 2018 as a spin-off from the University of St.Gallen, summarises, this would allow different and meaningful data to be collected and not mainly more of the same.

In addition, LightlyEdge also helps to find certain scenarios, which was previously like looking for a needle in a haystack in hours of videos. Thanks to advanced image-language embedding, it is possible to search for specific footage. The software marks the appropriate sequences according to textual search queries such as "ambulance with siren" or using the reference image of a snow-covered road at night.

The order "capture all frames that differ by at least 30 percent from what I have seen so far" is now also possible. Lightly Edge also provides insights into what devices see and thus, for example, into the operating environment of a fleet of delivery robots. ce/mm