NREC农业人员检测数据集
收藏资源简介:
NREC农业人员检测数据集是由卡内基梅隆大学国家机器人工程研究中心创建,用于推动非公路或农业环境中的人员检测研究。该数据集包含在橙园和苹果园中从两个感知平台(拖拉机和皮卡车)拍摄的标记立体视频,以及来自RTK GPS的车辆位置数据。数据集定义了一个基准,结合了总共76k标记的人员图像和19k采样的人员自由图像。该数据集突出了该领域的几个关键挑战,包括环境变化、植被遮挡、运动中的人员和非标准姿势,以及从各种距离看到的人员;元数据包括允许针对这些效果进行有针对性的评估。最后,我们展示了三种领先的城市行人检测方法和我们自己的卷积神经网络方法的基准检测性能结果,该方法受益于额外图像上下文的合并。我们表明,现有方法在城市数据上的成功并不直接转移到这一领域。
The NREC Agricultural Pedestrian Detection Dataset was developed by the National Robotics Engineering Center at Carnegie Mellon University to advance research on pedestrian detection in off-highway or agricultural environments. This dataset comprises annotated stereo videos captured from two perception platforms (tractor and pickup truck) in orange groves and apple orchards, alongside vehicle position data collected via RTK GPS. The dataset defines a benchmark that combines a total of 76k annotated pedestrian images and 19k sampled person-free images. This dataset highlights several core challenges in the field, including environmental variations, vegetation occlusion, moving pedestrians with non-standard postures, and pedestrians observed at varying distances; its metadata enables targeted evaluation of these specific effects. Finally, we present benchmark detection performance results for three state-of-the-art urban pedestrian detection methods and our own convolutional neural network (CNN)-based approach, which benefits from the incorporation of additional image context. We demonstrate that the success of existing methods on urban datasets does not directly transfer to this agricultural domain.

- 1Comparing Apples and Oranges: Off-Road Pedestrian Detection on the NREC Agricultural Person-Detection Dataset国家机器人工程研究中心,卡内基梅隆大学,匹兹堡,宾夕法尼亚州 15201 · 2017年



