[Seite 1]
The proposed 4D capture system is based on synchronised video cameras and light sources, control software, and high-resolution, software-based reconstruction methods for generating textured, watertight surface models arranged in continuous sequences, thereby enabling the capture of dynamic scenes. The optimal integration of hardware and software components into a unified system is required, encompassing control, synchronisation, calibration, data fusion, and analysis within a robust overall solution.
In this context, modularity, adaptability, and scalability for diverse projects are crucial. This applies specifically to the capture volume, the deployed acquisition hardware (including cameras, lighting equipment, and rigging), the control software (data storage and management), and the reconstruction software (both turnkey solutions and custom- built pipelines). Applications range from maximum reconstruction accuracy in record- then-reconstruct workflows to real-time analyses in interactive systems.
Key requirements:
Overall System: An optimal, ready-to-use system for the high-precision, markerless, 4- dimensional ("3D + time") capture and reconstruction of surfaces and object movements.
Cameras, Projectors, Illumination: The system shall consist of at least 70 synchronised cameras with individually selectable viewpoints as well as with camera resolution of at least 12 Megapixel (MP), ideally 12 Bit/pixel, potentially in a combination of greyscale and colour cameras to maximise SNR/accuracy while still capturing 3D- surface and texture. Cameras are synchronised, supplemented by eye-safe lighting and optional texture projectors. Alternatively, another camera count is appropriate, if equal or better capture performance is obtained compared to the described configuration. The system shall be expandable (cameras, illumination, etc.) through market-available components.
Measurement Volume/Coverage: Robust 360° capture (human full-body, including head, arms, legs, and feet); an acquisition volume of at least approx. 3 m × 3 m with a height of 2.5 m, providing a high viewpoint density to minimise occlusions. An extended capture volume of at least 5 m × 5 m × 2.5 m shall be configurable, at reduced accuracy and/or increased reconstruction deviation.
Total System Height: The overall height of the system must not exceed 3.1 m. Size of the room: 21m x 7m, usable 12m x 7m, height 3.1m at lowest point.
Geometric Accuracy: Dense surface reconstruction with deviations from the real geometry in the range of 0.5–1.0 mm; mean absolute deviation (MAD) ≤ 0.7 mm continuously across recordings (particularly for soft tissue deformations and per-vertex analyses). No frame drops.
Temporal Resolution: A minimum of 60 3D-fps (3D models per second), ideally 200 3D-fps, to capture rapid movements and deformations.
Video Sequence Length: Continuous recording of at least 8 minute video sequences to capture realistic human movement sequences.
[Seite 2]
Synchronisation: Hardware-based synchronisation with external sensors with an accuracy of at least 60 Hz (e.g., via PTP IEEE 1588 or a dedicated sync hub), for integration with systems such as IMUs, marker-based MoCap, or EMG. Optional: Real- time Ethernet via TSN (Time-Sensitive Networking) for data streaming.
Texture/Fidelity: High-quality RGB texture and detail reproduction to accurately capture wrinkling in textiles, fabric shifting, and material stretching (e.g., in human skin). Typical surfaces to be captured include textiles, human skin, and robots (potentially enhanced with additional texture, e.g., via texture spray).
Storage/4D Reconstruction/Processing: 19-inch rack-mountable GPU/CPU/disk components to provide uncompressed video data storage and reconstruction capacity to be located in a separate server room cabinet, cable distance ca. 40m. Uncompressed storage capacity for at least 5 recording sequences. Video sequence data will be deleted after reconstruction. Frame ranges shall be selectable before reconstruction, e.g., based on selected camera views. Storage to hold reconstructed 3D/4D data is not required within the requested system.
Software Integration: Uncompressed image/video output/API access is required to utilise market-available, distributed, GPU-based reconstruction software (e.g., RealityScan). The reconstruction must not rely on any assumptions about the objects to be reconstructed. For example, the use of an assumed human skeletal model or similar model is not permitted. The reconstruction software must run on processing components provided with the system. Capability to utilise additional GPU/CPU systems of the customer for reconstruction. Base system control software (e.g. for cameras, lighting, synchronisation) must have a perpetual license.
Open Data Formats: Export capabilities in standard formats (e.g., PLY, OBJ, STL), including time-coherent surface information, for seamless integration into machine learning (ML), simulation, and analysis pipelines.
Workflow and Reproducibility: Automated, batch-capable processing (calibration, registration, and consistent mesh generation), including quality control mechanisms to ensure the reproducible processing of large data volumes. Arrangement of cameras, lighting equipment, etc., adaptable to project needs with re-calibration procedure on non- reflective stand/rig.
Operational Safety and Support: The system must comply with all relevant safety standards. Requirements include CE conformity, reliable calibration, maintainable components, as well as manufacturer support and software updates. A breakdown of anticipated ongoing annual costs for maintenance and support must be provided.
Real-time Tracking Data (optional): Low-latency, Ethernet-based real-time streaming of tracking data from cameras (e.g., skeleton tracking with or without markers), potentially at a reduced data rate (less cameras or reduced image size).
Real-time Reconstruction (optional): Real-time streaming of segmented and unsegmented point clouds, potentially at a reduced resolution. Simultaneous recording of
[Seite 3]
uncompressed video for later reconstruction alongside reduced-quality real-time reconstruction and streaming.
Environmental Protection Requirements:
• Use of energy-efficient components. Automated standby management (for cameras and servers) and the capability to offload computationally intensive analyses to High-Performance Computing (HPC) infrastructure.
• Provisions for returning end-of-life equipment to certified recyclers, as well as the continued reuse of functional subcomponents.
Additional Requirements:
• Warranty: at least 24 months.
• Support: Response time < 24 hours (Monday to Friday).
• Delivery time below 6 months
• Documentation on system operation and maintenance, as well as training in system operation and maintenance, must be included in the tender.
• A breakdown of the anticipated recurring annual costs for maintenance and support must be provided.
• Costs for cameras, lenses, networking and connectivity equipment, computing and storage systems, installation and commissioning, and training must be itemised separately.