生活场景目标检测与抓取数据集

本数据集基于YOLOv5算法的目标检测训练需求构建，旨在为室内环境中常见物体识别提供高质量、可复现的视觉样本支持。数据采集时间为2024年6月至2025年5月，采集环境包括实验室与半自然家庭场景，涵盖多类生活物体与工具。图像数据由Intel RealSense D435i深度相机同步采集，彩色图像分辨率640×480，空间精度±2毫米。标注文件采用YOLO标准格式，记录目标类别编号与边界框归一化参数。为确保样本质量，数据采集过程中执行了光照、遮挡与相机姿态控制策略，数据处理阶段实施了重复检测去重与异常样本剔除机制，并通过Mosaic拼接、HSV扰动与仿射变换算法生成增强样本，增强参数固定以保证实验可复现性。本数据集的质量控制遵循国家科学数据管理要求，确保采集、加工、标注与验证的全流程一致性。该数据集可应用于目标检测、机器人视觉、场景识别等任务，具备良好的扩展性与学术研究价值。

This dataset is developed to meet the object detection training requirements of the YOLOv5 algorithm, aiming to provide high-quality and reproducible visual sample support for common object recognition in indoor environments. Data collection was conducted between June 2024 and May 2025. The acquisition environments cover laboratories and semi-natural household scenarios, including various daily objects and tools. Image data was synchronously captured using an Intel RealSense D435i depth camera, with color images having a resolution of 640×480 and a spatial accuracy of ±2 mm. Annotation files follow the YOLO standard format, recording target category IDs and normalized bounding box parameters. To guarantee sample quality, lighting, occlusion and camera pose control strategies were implemented during data collection. In the data processing phase, mechanisms for deduplicating duplicate detections and removing abnormal samples were adopted. Furthermore, augmented samples were generated via Mosaic stitching, HSV perturbation and affine transformation algorithms, with all augmentation parameters fixed to ensure experimental reproducibility. The quality control of this dataset complies with national scientific data management regulations, ensuring the consistency of the entire workflow covering collection, processing, annotation and verification. This dataset can be applied to tasks such as object detection, robotic vision and scene recognition, and exhibits excellent scalability and academic research value.

搜集汇总

背景与挑战

背景概述

该数据集是一个针对室内生活场景的目标检测与机器人抓取任务而构建的视觉数据集，基于YOLOv5算法需求设计，包含2024个文件（总大小680.11MB），采集于2024年6月至2025年5月，覆盖实验室和半自然家庭环境。数据由Intel RealSense D435i深度相机同步采集，提供640×480分辨率的彩色图像和YOLO格式的标注，并实施了光照控制、去重、增强等质量控制措施，确保数据的高质量和可复现性，适用于目标检测、机器人视觉等研究。