OCID (Object Clutter Indoor Dataset)

开发用于处理现实世界中物体的机器人感知系统需要根据预期的操作域仔细检查计算机视觉算法。这需要大量的地面实况数据来严格评估算法的性能。 Object Cluttered Indoor Dataset 是一个 RGBD 数据集，其中包含每个对象的逐点标记点云。数据是使用两个位于不同高度的 ASUS-PRO Xtion 相机捕获的。它捕获对象、背景、上下文、传感器到场景距离、视点角度和照明条件的各种设置。 OCID 的主要目的是允许在杂乱量增加的场景中系统地比较现有的对象分割方法。此外，OCID 还为其他视觉任务（如对象分类和识别）提供真实数据。

Developing robotic perception systems for real-world object manipulation requires careful validation of computer vision algorithms against their intended operational domains. This demands extensive ground truth data to rigorously evaluate algorithm performance. The Object Cluttered Indoor Dataset (OCID) is an RGBD dataset that provides per-point annotated point clouds for each object. The dataset was captured using two ASUS-PRO Xtion cameras positioned at different heights, covering diverse settings including objects, backgrounds, contexts, sensor-to-scene distances, viewpoint angles, and illumination conditions. The primary objective of OCID is to enable systematic comparison of existing object segmentation methods in scenes with increasing clutter levels. Furthermore, OCID also provides ground truth data for other computer vision tasks such as object classification and recognition.

搜集汇总

背景与挑战

背景概述

OCID是一个2019年发布的RGBD数据集，专为机器人感知系统设计，包含逐点标记的点云数据，使用双相机捕获以覆盖多样化的对象、背景和场景条件。其主要目的是在杂乱室内环境中系统评估对象分割方法，同时为对象分类和识别等视觉任务提供真实数据支持。