关于配置ORB-SLAM3

环境：VMware 17.3Pro Ubuntu22.04

一、配置ORB-SLAM3：

流程记录：

1.更换阿里云 APT 软件源，之后效果快得起飞

2. 第一阶段，有一条命令优化（某个apt库需要下载）

3. 编译安装Opencv：优化了命令配置 make -j4

4. ORB-SLAM3的编译，使用了12GB的分配+make -j2

参考：此博客

二、下载数据集，运行算法

5. TUM-VI数据集，我是通过WSL2终端命令下载的，之后复制到VMware。运行的时候需要修改源码来可视化，参考：此博客

6. 完整12GB的Euro数据集已经下载，同样是终端下载。这个数据集的运行效果不佳。

7. 命令：

   进入项目根目录

   cd ~/Dev/ORB_SLAM3

   确保程序有执行权限

   chmod +x ./Examples/Stereo/stereo_euroc

   Euroc MH01_easy 数据集:

   运行双目（完整运行）

   ./Examples/Stereo/stereo_euroc
   ./Vocabulary/ORBvoc.txt
   ./Examples/Stereo/EuRoC.yaml
   ~/Datasets/EuRoc/MH01
   ./Examples/Stereo/EuRoC_TimeStamps/MH01.txt
   dataset-MH01_stereo

   运行单目（无画面）

   ./Examples/Monocular/mono_euroc
   ./Vocabulary/ORBvoc.txt
   ./Examples/Monocular/EuRoC.yaml
   ~/Datasets/EuRoc/MH01
   ./Examples/Monocular/EuRoC_TimeStamps/MH01.txt
   dataset-MH01_mono

   运行单目+惯性传感器（初始化失败、卡顿）

   ./Examples/Monocular-Inertial/mono_inertial_euroc
   ./Vocabulary/ORBvoc.txt
   ./Examples/Monocular-Inertial/EuRoC.yaml
   ~/Datasets/EuRoc/MH01
   ./Examples/Monocular-Inertial/EuRoC_TimeStamps/MH01.txt
   dataset-MH01_monoi

   运行双目+惯性传感器（无画面）

   赋予执行权限

   chmod +x ./Examples/Stereo-Inertial/stereo_inertial_euroc

   运行命令

   ./Examples/Stereo-Inertial/stereo_inertial_euroc
   ./Vocabulary/ORBvoc.txt
   ./Examples/Stereo-Inertial/EuRoC.yaml
   ~/Datasets/EuRoc/MH01
   ./Examples/Stereo-Inertial/EuRoC_TimeStamps/MH01.txt
   dataset-MH01_stereoi

   TUM-VI数据集

   1. room1_512数据集

   单目：正常跑通

   chmod +x ./Examples/Monocular/mono_tum_vi

   ./Examples/Monocular/mono_tum_vi
   ./Vocabulary/ORBvoc.txt
   ./Examples/Monocular/TUM-VI.yaml
   ~/Datasets/TUM_VI/dataset-room1_512_16/mav0/cam0/data
   ./Examples/Monocular/TUM_TimeStamps/dataset-room1_512.txt
   dataset-room1_512_mono

   2. corridor1_512数据集

   单目：正常跑通

   ./Examples/Monocular/mono_tum_vi
   Vocabulary/ORBvoc.txt
   Examples/Monocular/TUM-VI.yaml
   ~/Datasets/TUM_VI/dataset-corridor1_512_16/mav0/cam0/data
   Examples/Monocular/TUM_TimeStamps/dataset-corridor1_512.txt
   dataset-corridor1_512_mono

   单目+惯性器：正常跑通

   ./Examples/Monocular-Inertial/mono_inertial_tum_vi
   Vocabulary/ORBvoc.txt
   Examples/Monocular-Inertial/TUM-VI.yaml
   ~/Datasets/TUM_VI/dataset-corridor1_512_16/mav0/cam0/data
   Examples/Monocular-Inertial/TUM_TimeStamps/dataset-corridor1_512.txt
   Examples/Monocular-Inertial/TUM_IMU/dataset-corridor1_512.txt
   dataset-corridor1_512_monoi

   双目+惯性：正常跑通

   ./Examples/Stereo-Inertial/stereo_inertial_tum_vi
   ./Vocabulary/ORBvoc.txt
   ./Examples/Stereo-Inertial/TUM-VI.yaml
   ~/Datasets/TUM_VI/dataset-corridor1_512_16/mav0/cam0/data
   ~/Datasets/TUM_VI/dataset-corridor1_512_16/mav0/cam1/data
   ./Examples/Stereo-Inertial/TUM_TimeStamps/dataset-corridor1_512.txt
   ./Examples/Stereo-Inertial/TUM_IMU/dataset-corridor1_512.txt
   dataset-corridor1_512_stereoi

三、EVO评估算法性能

8. 安装，参考：此博客

9. 评估：

(1) 这一部分我主要评估TUM-VI 的corridor1数据集的运行结果。

参考：此博客，也是遇到了同样的问题。

首先是安装缺失的包；

关闭纳秒解析：

evo_config set euroc_use_nanoseconds false

然后因为ORB-SLAM3 输出是 TUM 格式 (t, x, y, z, qx, qy, qz, qw)，TUM-VI data.csv 是 EuRoC 格式 (t, x, y, z, qw, qx, qy, qz)，编写脚本转化数据，转为秒、TUM格式：

import numpy as np

def fix_gt():
    with open('corridor_gt_raw.csv', 'r') as f:
        lines = f.readlines()
    with open('gt_final.tum', 'w') as f:
        for line in lines:
            if line.startswith('#'): continue
            p = line.replace(',', ' ').split()
            if len(p) < 8: continue
            # 将纳秒转为秒，保留 9 位小数
            ts = float(p[0]) / 1e9
            f.write(f"{ts:.9f} {p[1]} {p[2]} {p[3]} {p[5]} {p[6]} {p[7]} {p[4]}\n")

def fix_res():
    with open('f_dataset-corridor1_512_stereoi.txt', 'r') as f:
        lines = f.readlines()
    with open('res_final.tum', 'w') as f:
        for line in lines:
            p = line.split()
            if len(p) < 8: continue
            # 将纳秒转为秒，保留 9 位小数
            ts = float(p[0]) / 1e9
            f.write(f"{ts:.9f} {p[1]} {p[2]} {p[3]} {p[4]} {p[5]} {p[6]} {p[7]}\n")

if __name__ == "__main__":
    fix_gt()
    fix_res()
    print("转换完成：时间戳已转为秒（TUM标准格式）")

(2) 运行命令

# 绘制轨迹
evo_traj tum res_final.tum --ref=gt_final.tum -p --plot_mode xyz

# 计算APE
evo_ape tum gt_final.tum res_final.tum -va -p --align --t_max_diff 0.1

# 保存轨迹png
evo_traj tum res_final.tum --ref=gt_final.tum --align -p --save_plot ~/Datasets/Output/trajectory_comparison.png

# 保存APE结果图像
evo_ape tum gt_final.tum res_final.tum -va --align --t_max_diff 0.1 --save_plot ~/Datasets/Output/my_ape_result.png

# 保存RPE结果图像
evo_rpe tum gt_final.tum res_final.tum -va -p --align --t_max_diff 0.1 -d 1.0 -u m --save_plot ./rpe_result.png

# 保存APE的评估数据到zip
evo_ape tum gt_final.tum res_final.tum -va --align --t_max_diff 0.1 --save_results ~/Datasets/Output/stereo_inertial.zip

# 评估 RPE 并保存结果到 zip
evo_rpe tum gt_final.tum res_final.tum -va --align --t_max_diff 0.1 -d 1.0 -u m --save_results ~/Datasets/Output/stereo_inertial_rpe.zip