Orbbec EGO

Femto Bolt 双目鱼眼红外采集设备

分辨率1600x1300

帧率30 fps

IMU1000 Hz

设备SNAZER76400GZ

镜头Fisheye Stereo

畸变KannalaBrandt8

Orbbec_Ego_AZER76400GZ_20260806_150808

Recording #150808 Pipeline Report

2026-08-06 ~35.6s @ 30fps Orbbec Femto Bolt 1600x1300 Fisheye

1,040

Frames

Left+Right synced pairs

92.3

Contact Ratio (%)

960/1040 frames

100.0%

Quality Score (L5)

Verdict: PASS

管线状态

L1 Sync

L2 Pose

L3 Hands

L3b WiLoR

L3c HaWoR

L4 Semantic

L5 QC

EGO

采集画面预览

Orbbec EGO 双目 1600×1300 @30fps | 165° 鱼眼广角 | 全局快门

双目立体视差对

左目 + 右目同步帧 | 基线 ~65mm | ORB-SLAM3 立体惯导输入

左目 (参考相机)

右目

L1

时间同步质量

1,040 pairs

双目帧时间戳精确配对 | 左目参考 | IMU 1000 Hz

5.5

LR Mean Diff (μs)

max=79μs, std=10.5μs

33.333

帧间隔 (ms)

30fps target

1000

IMU 采样率 (Hz)

1039 segments

双目同步误差分布

LR diff histogram (1040 pairs)

帧间隔时序

参考线: 33.333ms

双目同步质量

同步帧对1,040 pairs

原始左目1040 frames

原始右目1059 frames (trimmed 0)

LR max diff79 μs

LR mean diff5.5 μs

LR std diff10.5 μs

IMU 集成

IMU 分段数1039

Accel/帧33.3

采样率1000 Hz

时移补偿-12282 μs

参考相机left_camera

L2

Camera Trajectory (ORB-SLAM3)

1040 poses

ORB-SLAM3 Stereo（纯双目，米制尺度来自 120 mm 基线） | KannalaBrandt8 鱼眼畸变模型 | 最多200个采样点

1,040

Pose Count

100.0% coverage (1040/1040)

1.1161

总路程 (m)

avg 0.0314 m/s

0.064 × 0.038 × 0.039

Bounding Box (m)

Stereo（120 mm 基线定尺度）

鼠标拖拽旋转 | 滚轮缩放 | 右键平移

轨迹统计

Pose count1,040

Coverage100.0% (1040/1040)

总路程1.1161 m

平均速度0.0314 m/s

Bounding Box0.064 × 0.038 × 0.039 m

MethodStereo（纯双目）

LensKannalaBrandt8 (Fisheye)

IMU 分段1039 segs

L3

手物交互检测 (Hands23)

92.3% contact

Hands23 Detector (NeurIPS 2023) | 逐帧手-物体交互状态、抓取类型

92.3%

接触帧占比

960 / 1040 frames

3

动作片段数

有效交互片段

3,125

原始手检测

ghost 1072 (34.3%) 已清理

Grasp 类型分布

Contact 分布

L3b

3D 手骨骼重建 (WiLoR)

3,306 hands

WiLoR-mini (CVPR 2025) | MANO 21 关键点 | 世界坐标系

3,306

Hands detected

across 1040 frames

3.2

Hands/Frame

avg detection count

21

Keypoints/Hand

3D world coords (63 dim)

骨骼图例左手 (L) 右手 (R) 腕部关键点输出: 21 关键点 (x,y,z) + 2D 投影坐标 / 帧

L3c

世界坐标手部轨迹 (HaWoR)

1039 frames

HaWoR (CVPR 2025) | 左右手腕 3D 世界坐标轨迹 | 最多200个采样点

150808 · Left & Right Wrist 3D Trajectory 1039 frames · 35.6s

图例: 左手轨迹右手轨迹地面网格拖拽旋转 · 滚轮缩放 · 右键平移

左手腕 (Left Wrist)

有效帧1039/1039

X range[0.058, 0.425] m

Y range[-0.749, -0.2] m

Z range[-0.77, -0.416] m

Mean pos(0.26, -0.52, -0.59)

右手腕 (Right Wrist)

有效帧1039/1039

X range[0.194, 0.525] m

Y range[-0.616, -0.074] m

Z range[-0.975, -0.434] m

Mean pos(0.34, -0.33, -0.57)

L4

语义标注 (gpt-4o)

18/18 ✓

关键帧采样 | 物体识别 + 动作分类 + 任务阶段

18

标注帧数

18/18 (100%)

13

物体类别

跨帧去重统计

4

任务阶段

holding, idle, manipulating, reaching

任务阶段分布

物体词表 (13 类)

coffee_cupcuphand_sanitizerlaptoporange_bottlepaper_cupphonesmartphonesnack_boxsnack_containersnack_packagetissue_boxwater_bottle

逐帧语义标注详情

18 帧 | gpt-4o

L5

质量验证

24 pass

质检合格该段已纳入数据集

100.0%

Quality Score

24 checks total — Verdict: PASS

✓L1file_exists:l1_sync.json

✓L1file_exists:imu_segments.json

✓L3file_exists:l3_actions.json

✓L4file_exists:l4_semantic.json

✓L2file_exists:l2_trajectory.json

✓L2file_exists:CameraTrajectory.txt

✓L1frame_count_positive1040 pairs

✓L1timestamps_monotonic

✓L1lr_sync_qualityp99=46μs, mean=5.5μs, max=79μs

✓L1frame_interval_stabilitymean=33.333ms, std=0.028ms

✓L1imu_segments_present1039 segments, 34643 accel samples

✓L1→L2pose_coverage99.9% of 1040 frames have a valid camera pose

✓L1→L2trajectory_reasonabledist=1.1161m, bbox=[0.0639, 0.0385, 0.0394], speed=0.0314m/s

✓L1→L3frame_count_match1040 frames

✓L1→L3frame_idx_alignment1040 indices aligned

✓L3contact_ratio_plausible92.3%

✓L3hand_count_rangemax=2 hands/frame (clean)

✓L3action_segments_valid3 segments, all ≥0.3s

✓L3postprocessing_appliedghost_removal=34.3%, smoothing=0%

✓L3→L4keyframe_indices_in_l3All 18 keyframes have L3 data

✓L4annotation_completeness18/18 (100%)

✓L4annotation_schema_validAll 18 annotations have required fields

✓L4objects_detected13 unique object categories

✓L4phase_diversity4 phases: ['holding', 'idle', 'manipulating', 'reaching']

所有检查通过。L1-L4 层质量验证完成。

L0-L6

数据处理管线

L1-L5 Complete

从原始双目采集到具身智能训练数据的完整处理链路

L1

Frame Sync

✓ Complete

L2

ORB-SLAM3

✓ Complete

L3

Hands23

✓ Complete

L3b

WiLoR

✓ Complete

L3c

HaWoR

✓ Complete

L4

GPT-4.1

✓ Complete

L5

QC 100.0%

✓ Complete

L6

LeRobot v3

✓ Complete

DONE

Export

✓ Complete