← 返回 Skills 市场

D455相机数据采集工具

Name: D455相机数据采集工具
Author: qujingyang28

作者 Robot_Qu · GitHub ↗ · v1.0.0 · MIT-0

cross-platform ✓ 安全检测通过

100

总下载

当前安装

版本数

在 OpenClaw 中安装

/install box-pointcloud-collector

功能描述

Intel RealSense D455 箱体点云数据采集工具。用于采集箱体的 RGB 彩色图和深度图数据集，支持手动/自动采集模式。触发场景：(1) 采集箱体/物体点云数据 (2) 创建深度学习训练数据集 (3) Bin Picking 项目数据准备 (4) 相机标定数据采集。支持 Intel RealSens...

使用说明 (SKILL.md)

箱体点云数据采集工具

使用 Intel RealSense D455 相机采集箱体 RGB-D 数据集。

快速开始

启动采集

python scripts/collect_v2.py

操作按键

按键	功能
S	保存 1 张（RGB + Depth）
A	自动保存 10 张（每 5 秒 1 张）
Q	退出并打开数据目录

数据集结构

box_dataset/
└── box_YYYYMMDD_HHMMSS/
    ├── camera_intrinsic.json    # 相机内参
    ├── dataset_config.json      # 数据集配置
    ├── capture_stats.json       # 采集统计
    └── raw_data/
        ├── 000001_Color.png     # RGB 彩色图
        ├── 000001_Depth.raw     # 深度图 (uint16)
        └── ...

文件格式

文件	格式	说明
`XXX_Color.png`	PNG 8位 RGB	彩色图像
`XXX_Depth.raw`	uint16 raw	深度数据 (640×480×2 bytes)
`camera_intrinsic.json`	JSON	相机内参 (fx, fy, cx, cy)

数据处理

读取深度图

import numpy as np

depth = np.fromfile("000001_Depth.raw", dtype=np.uint16)
depth = depth.reshape(480, 640)
depth_m = depth / 1000.0  # 转换为米

像素坐标 → 3D 坐标

import json

with open("camera_intrinsic.json") as f:
    intr = json.load(f)

def pixel_to_3d(u, v, depth_m):
    X = (u - intr['ppx']) * depth_m / intr['fx']
    Y = (v - intr['ppy']) * depth_m / intr['fy']
    Z = depth_m
    return X, Y, Z

采集规范

有效距离：0.5m - 2.5m
推荐距离：0.5m - 1.5m
每种箱体：20-30 张
变换角度：调整箱子位置/角度增加多样性

注意事项

中文路径问题：数据集目录使用纯英文命名
键盘焦点：点击 RGB Camera 窗口后按 S/A 键
相机占用：关闭 RealSense Viewer 等程序

硬件要求

Intel RealSense D455
USB 3.0 接口
Windows 10/11

依赖安装

pip install pyrealsense2 opencv-python numpy

作者：Robot_Qu
版本：v2.0

安全使用建议

This skill appears coherent and local-only: it captures RGB/D frames from a RealSense D455 and saves them for processing. Before running, verify you are on Windows with a connected D455 and have installed pyrealsense2, opencv-python, and numpy. Review and (preferably) change the BASE_DIR constant in scripts/collect_v2.py to a directory appropriate for your user (it currently targets C:\Users\Administrator\.openclaw\workspace\box_dataset). Expect files to be written there and the folder to be opened automatically (os.startfile). There are no network calls or credential requests in the code, so the primary risks are accidental overwriting of files, permission errors on non-Administrator accounts, and the usual safety considerations when running third-party scripts locally.

功能分析

Type: OpenClaw Skill Name: box-pointcloud-collector Version: 1.0.0 The skill bundle is a legitimate data collection tool for Intel RealSense D455 cameras, designed to capture RGB-D datasets for box detection. The scripts (scripts/collect_v2.py and scripts/process_data.py) perform standard hardware interfacing, image processing, and local file I/O within a designated workspace directory, with no evidence of network exfiltration, credential theft, or malicious prompt injection.

能力评估

✓ Purpose & Capability

Name/description match the provided scripts and docs: collect_v2.py captures RGB and depth frames from an Intel RealSense D455, saves raw files and metadata, and process_data.py demonstrates simple processing. No unrelated capabilities (cloud, remote APIs, or extra service credentials) are requested.

ℹ Instruction Scope

SKILL.md instructs running the included scripts and installing pyrealsense2/opencv-python/numpy — that aligns with the purpose. The instructions and code only read/write local files and the camera device. Note: the scripts use a hardcoded Windows path (C:\Users\Administrator\.openclaw\workspace\box_dataset) and call os.startfile(), so they assume a Windows environment and will create files under an Administrator-profile path unless you edit the BASE_DIR.

✓ Install Mechanism

No install spec; dependency installation is via pip (recommended in SKILL.md). No downloaded archives, no external installers, and no execution of remote code beyond installing standard Python packages.

ℹ Credentials

The skill requests no environment variables or credentials, which is proportionate. The only notable environmental assumption is the hardcoded BASE_DIR under an Administrator Windows home; this is not a secret access request but may be inappropriate for non-Administrator or non-Windows systems and could cause permission or path issues.

✓ Persistence & Privilege

always is false and the skill does not request elevated or persistent platform privileges. It writes data only to its own dataset directory and does not modify other skill configs or global system settings.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install box-pointcloud-collector
安装完成后，直接呼叫该 Skill 的名称或使用 /box-pointcloud-collector 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.0

- Initial release of box-pointcloud-collector. - Collects RGB and depth dataset of boxes using Intel RealSense D455 camera. - Supports manual and automatic data collection modes. - Provides a structured data output and sample code for data reading and processing. - Designed for bin picking, deep learning dataset creation, and camera calibration scenarios.

元数据

Slug box-pointcloud-collector

版本 1.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 1

常见问题