Novel Dataset of RGB-D Images for Robotic Harvesting of Green Asparagus

Enabling robotic harvesting of green asparagus requires the robust detection and localization of asparagus spears in the field. The dataset contains typical environmental measurement impact, such as a variety of natural spear patterns, weeds, natural lighting, and the effect of ground irregularities. From 7 green asparagus sites located in Germany and France approximately 20 m of one bed of each site were captured with a camera in top down position, perpendicular to the ground. The dataset contains RGB images and corresponding depth recorded from the video stream of a IntelRealsense D405 camera. The images have 1280 x 720 pixel resolution and 0.1 mm depth resolution. The average ground distance is around 500 mm. Driving speed around 300 mm/s or above and a frame rate of 30 fps result in approximately 1800 RGB and depth image pairs per site. RGB images are provided in CV_8UC3 png format, the depth map in CV_16UC1 png format.The file readme.txt shares details of the folder structure and file formats.