目录

1 寻找目标领域数据集并下载

2 数据集处理及运行

2.1 单目

2.2.1 图片

2.2.2 时间戳

2.2.3 相机参数yaml编写

2.2.4 运行测试

2.2 双目

2.2.1 图片

2.2.2 时间戳

2.2.3 相机参数yaml编写

2.2.5 运行测试

2.3 RGB-D

---

SLAM进阶专栏部分，系统梳理了SLAM各环节的主流算法、在具身智能与大模型中的主流使用方法，配合博主自行编写的demo进行效果展示。该专栏旨在细节了解SLAM技术在具身智能领域的主流算法、及大模型对于SLAM技术的赋能。

1 寻找目标领域数据集并下载

（1）TUM数据集链接：https://cvg.cit.tum.de/data/datasets/rgbd-dataset/download

（2）KITTI数据集链接：https://www.cvlibs.net/datasets/kitti/eval_odometry.php

（3）EuRoC数据集链接：https://projects.asl.ethz.ch/datasets/doku.php?id=kmavvisualinertialdatasets

（4）FinnForest数据集链接：

https://etsin.fairdata.fi/datasets

（5）可以根据自己领域制作/寻找

2 数据集处理及运行

2.1 单目

基于ORB_SLAM2运行，执行文件为mono_tum.cc

2.2.1 图片

文件名rgb要求格式

000000.png
000001.png
000002.png
...

图片放置路径：/media/ch/WD_BLACK/dataset/S01_13Hz_summer_seq1_shortLoop/S01_13Hz/rgb

2.2.2 时间戳

文件名rgb.txt，时间戳格式要求：

0.000000 rgb/000000.png
0.060000 rgb/000001.png
0.140000 rgb/000002.png
0.220000 rgb/000003.png
0.300000 rgb/000004.png
0.360000 rgb/000005.png
0.440000 rgb/000006.png
...

（1）拍摄时间格式

0.000000
0.060000
0.140000
0.220000
...

a.创建文本orb_timestamps.py

b.调整为图片标准形式+时间，需要写个脚本进行转换

import os

# 时间戳文件路径
TIMESTAMP_FILE = 'times_S01.txt' 
# 图像子文件夹名
IMAGE_DIR = 'rgb' 
# 输出文件名
OUTPUT_FILE = 'rgb.txt' 
# -----------------------------------

try:
    with open(TIMESTAMP_FILE, 'r') as f:
        timestamps = [line.strip() for line in f if line.strip()]
except FileNotFoundError:
    print(f"错误：未找到文件 {TIMESTAMP_FILE}，请检查路径。")
    exit()

num_images = len(timestamps)

with open(OUTPUT_FILE, 'w') as f:
    for i in range(num_images):
        # 000000.png, 000001.png 等文件名
        filename = f'{i:06d}.png' 
        
        # 写入一行: 时间戳 文件路径/文件名
        timestamp = timestamps[i]
        line = f'{timestamp} {IMAGE_DIR}/{filename}\n'
        f.write(line)

print(f"成功生成 {num_images} 个关联，保存到 {OUTPUT_FILE}。")

c.运行指令，生成orb_timestamps.py文件

python3 orb_timestamps.py

d.检查

e.时间戳路径：rgb.txt

2.2.3 相机参数yaml编写

参数需要根据实际采集的数据集拍摄相机进行修改，yaml配置文件名Finn_s01.yaml

%YAML:1.0

#--------------------------------------------------------------------------------------------
# Camera Parameters. Adjust them!
#--------------------------------------------------------------------------------------------

# Camera calibration and distortion parameters (OpenCV) 
Camera.fx: 1056.6287421208997
Camera.fy: 1056.9497022783046
Camera.cx: 952.1135175072209
Camera.cy: 592.824593628865

Camera.k1: -0.12427059109969174
Camera.k2: 0.08183525905697533
Camera.p1: 0.0
Camera.p2: 0.0
Camera.k3: -0.020818847104251116

# Camera frames per second 
Camera.fps: 13.0

# Color order of the images (0: BGR, 1: RGB. It is ignored if images are grayscale)
Camera.RGB: 1

#--------------------------------------------------------------------------------------------
# ORB Parameters
#--------------------------------------------------------------------------------------------

# ORB Extractor: Number of features per image
ORBextractor.nFeatures: 4000

# ORB Extractor: Scale factor between levels in the scale pyramid 	
ORBextractor.scaleFactor: 1.2

# ORB Extractor: Number of levels in the scale pyramid	
ORBextractor.nLevels: 8

# ORB Extractor: Fast threshold
# Image is divided in a grid. At each cell FAST are extracted imposing a minimum response.
# Firstly we impose iniThFAST. If no corners are detected we impose a lower value minThFAST
# You can lower these values if your images have low contrast			
ORBextractor.iniThFAST: 10
ORBextractor.minThFAST: 3

LoopClosing.minCovisible: 4
LoopClosing.maxDistance: 50

#--------------------------------------------------------------------------------------------
# Viewer Parameters
#--------------------------------------------------------------------------------------------
Viewer.KeyFrameSize: 0.05
Viewer.KeyFrameLineWidth: 1
Viewer.GraphLineWidth: 0.9
Viewer.PointSize:2
Viewer.CameraSize: 0.08
Viewer.CameraLineWidth: 3
Viewer.ViewpointX: 0
Viewer.ViewpointY: -0.7
Viewer.ViewpointZ: -1.8
Viewer.ViewpointF: 500

yaml路径：ORB_SLAM2/Examples/Monocular/Finn_s01.yaml

2.2.4 运行测试

cd ORB_SLAM2
./Examples/Monocular/mono_tum Vocabulary/ORBvoc.txt Examples/Monocular/Finn_s01.yaml /media/ch/WD_BLACK/dataset/S01_13Hz_summer_seq1_shortLoop/S01_13Hz/

2.2 双目

2.2.1 图片

需要左、右目数据集，根据stereo_kitti.cc启动文件将左目文件夹名称改为“image_0”、右目“image_1”，左右目中图片命名格式如下:

000000.png
000001.png
000002.png
...

2.2.2 时间戳

（1）根据stereo_kitti.cc启动文件，时间戳文本名称为“times.txt”

（2）时间戳文本内容格式

0.000000 image_0/000000.png image_1/000000.png
0.060000 image_0/000001.png image_1/000001.png
0.140000 image_0/000002.png image_1/000002.png

（3）数据集时间戳格式

0.000000
0.060000
0.140000

（4）左右目图片格式

（5）编写时间戳脚本

import os

# 输入文件和目录名
timestamp_file = 'times_S01.txt'  # 替换成您的原始时间戳文件名
output_file = 'timestamp_stereo.txt'
left_dir = 'image_0'
right_dir = 'image_1'
image_ext = '.png' # 图像文件扩展名

# 读取原始时间戳
with open(timestamp_file, 'r') as f:
    timestamps = [line.strip().split()[0] for line in f if line.strip()]

# 生成双目关联文件内容
output_lines = []
num_images = len(timestamps)

for i in range(num_images):
    # 确保图像文件名有足够的零填充 (6位)
    image_name = '{:06d}{}'.format(i, image_ext)

    left_path = os.path.join(left_dir, image_name)
    right_path = os.path.join(right_dir, image_name)

    # 格式：timestamp left_path right_path
    line = f"{timestamps[i]} {left_path} {right_path}\n"
    output_lines.append(line)

# 写入新的双目时间戳文件
with open(output_file, 'w') as f:
    f.writelines(output_lines)

print(f"成功生成双目时间戳文件：{output_file}，共 {num_images} 行。")

（6）生成目标时间戳文本times.txt，generate_stereo_timestamps.py为脚本文件名

python3 generate_stereo_timestamps.py

2.2.3 相机参数yaml编写

注意替换相机参数，视觉参数viewer parameters可以不改。

%YAML:1.0

#--------------------------------------------------------------------------------------------
# Camera Parameters. Adjust them!
#--------------------------------------------------------------------------------------------

# Camera calibration and distortion parameters (OpenCV) 
Camera.fx: 1056.6287421208997
Camera.fy: 1056.9497022783046
Camera.cx: 952.1135175072209
Camera.cy: 592.824593628865

Camera.k1: -0.12427059109969174
Camera.k2: 0.08183525905697533
Camera.p1: 0.0
Camera.p2: 0.0
Camera.k3: -0.020818847104251116

Camera.width: 2013
Camera.height: 1195

# Camera frames per second 
Camera.fps: 13.0

# stereo baseline times fx
Camera.bf: 211.77668

# Color order of the images (0: BGR, 1: RGB. It is ignored if images are grayscale)
Camera.RGB: 1

# Close/Far threshold. Baseline times.
ThDepth: 35

#--------------------------------------------------------------------------------------------
# ORB Parameters
#--------------------------------------------------------------------------------------------

# ORB Extractor: Number of features per image
ORBextractor.nFeatures: 4000

# ORB Extractor: Scale factor between levels in the scale pyramid 	
ORBextractor.scaleFactor: 1.2

# ORB Extractor: Number of levels in the scale pyramid	
ORBextractor.nLevels: 8

# ORB Extractor: Fast threshold
# Image is divided in a grid. At each cell FAST are extracted imposing a minimum response.
# Firstly we impose iniThFAST. If no corners are detected we impose a lower value minThFAST
# You can lower these values if your images have low contrast			
ORBextractor.iniThFAST: 20
ORBextractor.minThFAST: 5

#--------------------------------------------------------------------------------------------
# Viewer Parameters
#--------------------------------------------------------------------------------------------
Viewer.KeyFrameSize: 0.6
Viewer.KeyFrameLineWidth: 2
Viewer.GraphLineWidth: 1
Viewer.PointSize:2
Viewer.CameraSize: 0.7
Viewer.CameraLineWidth: 3
Viewer.ViewpointX: 0
Viewer.ViewpointY: -100
Viewer.ViewpointZ: -0.1
Viewer.ViewpointF: 2000

2.2.5 运行测试

./Examples/Stereo/stereo_kitti ./Vocabulary/ORBvoc.txt ./Examples/Stereo/Finn_stereo.yaml /media/ch/WD_BLACK/dataset/S01_13Hz_summer_seq1_shortLoop/S01_13Hz/

2.3 RGB-D

需要将RGB图片转化为png/jpg格式，深度图转为16bit-png（单位毫米），RGB-深度时间戳对齐。

指令格式：RGB-D可执行路径+词袋路径+配置文件路径+RGB文件路径+RGB-深度对齐文件路径，示例如下：