Name: TTok Vision Ego Data Capture Device Family
Brand: Tok Vision

Capture Portfolio

Multiple devices. One FPGA synchronization core.

From rapid algorithm validation to complex teleoperation demonstrations — every device form factor is synchronized by the same FPGA timing core, in hardware, before a single frame reaches the SoC.

Stereo USB

Stereo USB Capture Device

The lightweight entry point for research validation and algorithm iteration.

Driver-free USB 3.2 Gen1 (5 Gbps) — works with NVIDIA AI platforms
Synchronized stereo output for research pipelines
Desktop deployment — device to data in minutes

Stereo Vision

Standalone Stereo Vision

Stable stereo time-series data for spatial perception, localization, and mapping.

Independent on-device operation
Built for high-precision VIO, SLAM, and stereo algorithms
Hardware-level simultaneous stereo exposure

Multi-Camera

6-Camera Rig

Multi-view coverage that increases the information density of complex environments.

Unified clock drives coordinated multi-camera capture
Multi-sensor sync < 1 µs across all cameras (ns-class FPGA)
Inter-device multi-camera sync < 10 µs
Scalable to 2 / 4 / 6 / 8 camera configurations

Hand & Wrist

UMI Hand & Wrist Capture

Turns human demonstrations into trainable data for manipulation learning.

Records hand actions and wrist-perspective video
Sensor sync for wrist / UMI modules
Feeds grasping, assembly, and interaction policy training

FPGA ISP pipeline architecture: MIPI, LVDS and DVP input, RAW domain processing (DPC, BLC, LSC, HDR Fusion, AWB Gain, Joint DRC, Bayer de-mosaic RGGB), RGB domain (Gamma, CCM), YUV domain (CSC chroma processing, Contrast and Brightness, Edge Enhance, NR Luma), and outputs to Multiple V-Out, DMA Writer, Panning and Scaler, Memory Buffer, Display DPU — FPGA ISP Pipeline Architecture — production pipeline · 100% in-house IP · HDR up to 140 dB · industry-leading latency

Industrial Design

Ultra-light. Head, wrist, or chest.

Minimal, flexible industrial design with full customization support. Capture unit weighs under 120 g including the head strap — verified at 112.9 g on the bench.

Tok Vision ego capture device — front, side, and rear views

Data Modeling & Simulation

Validated in SLAM / VIO pipelines.

Every capture unit is evaluated in a stereo VIO pipeline before delivery — trajectory, feature tracks, and loop closure quality verified against grading thresholds.

Stereo VIO — live trajectory replay (RViz)

OpenVINS Data Evaluation Report

Auto-generated 2026-09-16 17:21:19

Grading — loop error rate = end offset / total path

S < 0.5% A < 1% B < 2% C < 3% D < 5% E ≥ 5%

File	Cameras	Init	Frames	Total Path	End Offset	Error Rate	Grade	Conclusion
run.log	2	STATIC	627	8.86 m	0.065 m	0.73 %	A	Grade A (error rate 0.73%)

Details — run.log Grade A

Cameras: 2

Init: STATIC

Frames: 627

Total path: 8.86 m

End position: [-0.03, -0.042, -0.039]

Loop closure error: 0.0647 m (0.73%)

Trajectory (XY top view)

Cumulative distance dist(t)

Diagnosis: no anomalies.

OpenVINS data evaluation report — Grade A, loop error rate 0.73% (auto-generated)

Stereo Camera Verification Report

Verified 2026-08-18 11:36:21 · calibration file stereo_calib_20260818_103124-YT7760.yaml · 1920 × 1200 · baseline 64.56 mm (measured)

CALIBRATION PASS PARAMETER CONSISTENCY ✓ RECTIFICATION PASS

① Calibration accuracy

Stereo RMS: 0.286 px

Quality grade: A (excellent) PASS

② Parameter consistency

Measured RMS: 0.264 px vs file: 0.286 px

Consistent poses: 7 / 7 CONSISTENT ✓

③ Rectification — epipolar error

Mean: 0.208 px · p95: 0.505 px

RMS: 0.258 px · max: 0.712 px

Valid corners: 616 · outliers: 0 · poses: 7 PASS

Per-pose rectification quality

Sample	mean (px)	95% (px)	RMS (px)	max (px)	Corners
Pose 1	0.216	0.500	0.264	0.630	88
Pose 2	0.254	0.576	0.311	0.699	88
Pose 3	0.224	0.453	0.267	0.623	88
Pose 4	0.194	0.463	0.242	0.657	88
Pose 5	0.212	0.507	0.257	0.657	88
Pose 6	0.188	0.451	0.238	0.712	88
Pose 7	0.170	0.415	0.213	0.572	88

Auto-generated by StereoVerifyTool

Per-unit factory verification — calibration, parameter consistency, and rectification all PASS before shipment

Specifications

Engineered for reproducible research.

Output follows top-tier conference ego dataset standards — experiments are easy to reproduce, and effective data yield is higher.

Category	Specification
Image Sensors	Stereo RGB global shutter
Resolution	1920×1200 @ 30/60 fps · 1920×1080 @ 30 fps · 1600×1200 @ 30 fps (4:3)
Field of View	H 160° / V 130° standard (multiple lens options, customizable)
Stereo Baseline	65 mm nominal (64.56 mm measured on the verification unit); rigid PCB + bracket construction
Multi-Sensor Sync	< 1 µs, ns-class hardware sync across all on-device sensors (FPGA); hardware-level simultaneous exposure
Inter-Device Multi-Camera Sync	< 10 µs
IMU	6-axis / 9-axis (TDK / ST), 300–600 Hz sampling
Camera–IMU Sync	Frame-to-IMU sync accuracy < 20 µs, hardware timestamping
Device Sync	PTP wired µs-class · NTP wireless ms-class · RF supported
Video Encoding	H.264 / H.265, bitrate 6–10 Mbps (adjustable)
Output Formats	MP4 / MCAP container; WAV / AAC audio (256 kbps, 16 kHz); dual video streams
Recording	8-hour continuous capture, zero dropped frames
Storage	64 GB onboard eMMC/SPI + microSD card up to 256 GB; 1 GB DDR memory
Connectivity	USB 3.2 Gen1 (5 Gbps, driver-free) · Wi-Fi · Gigabit Ethernet (RTSP)
Power	5 V USB Type-C (supports simultaneous data output), optional 18650 quick-swap battery compartment
Wearable Design	< 120 g including head strap (112.9 g verified); head, wrist, and chest mounting
Factory Calibration	Per-unit calibration at the factory: camera intrinsics, stereo extrinsics, IMU–camera extrinsics; calibration files bound to the device serial number
OTA Updates	Over-the-air firmware update supported (microSD card priority)
Privacy & Compliance	Recording indicator LED + camera status LED; 1-second long-press start/stop capture with onboard or external button pad
Adjustable Parameters	AE, AWB, exposure time, frame rate, contrast, and brightness

Warm-Storage NAS

Turn high-frequency capture into reliable warm-tier storage.

Our NAS product line gives capture devices a steady ingest path — stable throughput, scalable capacity, and RAID-protected data — so recordings land safely and stay ready for algorithm ingestion.

Topology 01

Point-to-Point

One capture device to one storage node. Direct deployment, one clean link.

Topology 02

Multipoint-to-Point

Multiple capture devices converge on one node — built for production-scale data management.

Topology 03

Storage & Protection

3.5-inch drive compatible, concurrent read/write, RAID backup for continuous ingest.

01CaptureStereo / multi-camera / UMI

02High-Speed WriteConcurrent ingest & aggregation

03Warm StorageNAS / RAID data protection

04AI GovernanceCleaning · QC · traceability

Automated AI cleaning, quality inspection, and governance upgrade your data from "stored" to trainable, manageable, and traceable.

Machine Vision Customization

A full-stack vision team behind your product.

We are an original source provider of vision products and technology — from self-developed ISP IP in silicon to complete camera systems, optics, algorithms, and hardware, optimized end to end.

ISP IP

Self-Developed FPGA ISP

A proprietary, highly customizable ISP pipeline designed for low-latency, high-synchronization scenarios — from RAW domain processing to YUV output.

Mass-produced and licensed to chipset vendors
HDR up to 140 dB; ultra-low latency, industry-leading
RGB / RCCB CFA support, LSC, AWB, demosaic, LDC

Ego ISP

Dedicated Ego-Capture IP

Purpose-built ISP IP for embodied data capture with sub-µs synchronization — a capability we believe leads the industry.

FPGA-side sync and alignment before encoding
Objective metrics for subjective image quality — an industry first
Low-light full-color technology

Engineering

Custom Development & Delivery

Complete mechanical, optical, and simulation design capability for full product-line customization.

SDK + calibration toolchain for secondary development
Custom recording hosts, data inspection & cleaning services
R&D → production → delivery of full ego capture devices

Live Algorithm

Visible-light deep dehazing.

Real-time DEHAZE enhancement running in our ISP pipeline — scene contrast and color fidelity recovered in real time, with no post-processing required.

Helmet-mounted capture device — CAD structural design with FOV rays — Helmet-mounted rig — structural design with FOV ray simulation

Cap-mounted capture device — custom housing design — Cap-mounted wearable form factor — custom housing design

About Tok Vision

Standards-driven. Full-stack. Delivery-proven.

Tok Vision is an original‑source provider of vision products and technologies. We run China’s only open‑access imaging laboratory and provide services to globally leading image quality teams.

IEEE P2020 Contributor

Active in 4 of 7 sub-working groups alongside ~300 global experts; one of the few voting members worldwide.

Deep IP Portfolio

50+ IP filings, in which invention patents account for 58% of the total. This includes 25+ granted patents and 15 software copyright registrations, focused on core imaging domains.

iWayLab

China's only open imaging laboratory — spectrum-programmable flat-light arrays, real vehicle head- and taillights, standard charts, and real-world spectral data from over 2M+ km of driving, powering subjective and objective testing.