Valeo has launched WoodScape*, the first-ever surround‑view fisheye camera open-source dataset, with the aim of taking research in computer vision for automated driving and parking to the next level.

“With WoodScape, we would like to encourage the research community to develop computer vision algorithms for fisheye cameras aimed at both low- and high-speed vehicle automation scenarios, in order to make roads safer”, said Marc Vrecko, President of Valeo’s Comfort & Driving Assistance Business Group.

WoodScape comprises over 10,000 images captured by multiple Valeo vehicles across Europe using four automotive-grade surround-view fisheye cameras and annotations for a variety of tasks such as semantic segmentation, depth estimation, 2D object detection, visual odometry, motion segmentation, soiling detection and end-to-end driving.

While datasets from narrow field cameras are available, there are no extensive multitask surround-view fisheye camera datasets publicly available. In addition, most automotive public datasets limit research to three or four tasks.

One of the biggest challenges in automated driving research is obtaining sufficient relevant data that has been thoroughly labelled and human-annotated to allow perception systems to classify objects surrounding the vehicle. WoodScape will help to address this challenge.

Valeo has developed the most comprehensive sensor portfolio in the automotive industry, comprising ultrasonic sensors, radars, cameras, and the first and only series-produced LiDAR to meet automotive standards and able to be fitted in vehicles that are already on the market. In particular, Valeo produces fisheye cameras enabling 360° perception, key for automated driving.

Source: Valeo