Machine Vision Camera OEM Manufacturer China

Sensor Selection — Global Shutter vs Rolling Shutter

The choice between global shutter and rolling shutter is not aesthetic — it is functional. In a rolling shutter sensor, rows are exposed sequentially from top to bottom. At a conveyor belt running at 1 m/s with a 10 ms exposure, a 50 mm-tall object produces roughly 50 µm of geometric distortion per row. For label printing inspection or PCB solder joint detection, that distortion is enough to shift a feature outside its acceptance window and generate false rejects. For moving-parts inspection of any kind, global shutter is mandatory.

The dominant global shutter sensor family for Chinese industrial camera OEMs is Sony Pregius S: IMX174 (2.3 MP, 5.86 µm pixel), IMX250 (5 MP, 3.45 µm pixel), IMX255 (8.9 MP, 3.45 µm pixel), and the newer IMX530 (24.5 MP, 2.74 µm pixel). All four are stacked CMOS sensors with on-chip ADC, which lets them deliver low readout noise at high frame rates — the IMX174 hits 163 fps at full resolution over GigE, the IMX250 sustains 75 fps. Sony’s BSI (backside illumination) architecture on the Pregius S line also improves quantum efficiency at 550–700 nm compared to the previous Pregius generation, which matters for defect detection under white LED ring lights.

Sony Starvis 2 (e.g. IMX585, IMX662) uses rolling shutter but offers substantially higher sensitivity — quantum efficiency peaks around 80% at 520 nm compared to ~70% for Pregius S. These sensors are the right choice for low-light static inspection: semiconductor die photography, dark-field surface inspection, or astronomy-adjacent scientific imaging. Never specify Starvis 2 for a moving-object line unless you have strobed illumination short enough to freeze motion, which in practice means pulse widths below 50 µs at typical conveyor speeds.

Pixel size trade-off. Larger pixels collect more photons per exposure, improving dynamic range and signal-to-noise ratio. The IMX174’s 5.86 µm pixel gives it 73–80 dB dynamic range — sufficient for most industrial contrast tasks. Smaller pixels (3.45 µm on IMX250/IMX255) pack more resolution onto the same sensor area at the cost of sensitivity per pixel. For applications where spatial resolution is the constraint — fine-pitch connector inspection, sub-100 µm feature detection — the 5 MP or 8.9 MP sensor at 3.45 µm is appropriate. For applications limited by lighting power or requiring fast shutter speeds in dim conditions, the larger pixel wins.

Monochrome vs colour. Monochrome cameras deliver roughly 3× the sensitivity of equivalent colour cameras at the same resolution, because there is no Bayer colour filter array attenuating incoming light. For defect inspection tasks where colour is not the discriminator — scratches, dimensional measurement, solder joint geometry — monochrome is the correct choice and the default for industrial lines. Colour cameras are appropriate when the defect itself is colour-coded: incoming inspection for packaging print, colour consistency for painted surfaces, or LED wavelength binning.

CMOS vs CCD. CCD sensors are legacy at this point. CMOS dominates area scan industrial cameras above 1 MP because of higher frame rates, lower power consumption, and better integration with on-chip processing. CCD still appears in niche hyperspectral and scientific cameras where its specific charge-transfer properties matter. For a standard industrial IoT machine vision deployment, CCD is not a realistic consideration.

Interface Protocol and Integration

GigE Vision (IEEE 802.3 Gigabit Ethernet) is the dominant interface for area scan cameras in factory environments. The practical advantages are well-understood: standard Cat6 cable runs up to 100 m without repeaters; Power over Ethernet (PoE, 802.3af/at) eliminates separate power cabling on the camera side; managed switches enable multi-camera synchronisation across a production line without additional synchronisation hardware; and the network topology scales from a single camera to dozens without changing the host interface. At 1 GbE bandwidth, a full-resolution IMX174 frame (2.3 MP × 8-bit mono = 2.3 MB) transfers in roughly 18 ms — viable for 50 fps sustained throughput with buffering. The newer 2.5GbE GigE Vision cameras address the bandwidth ceiling and are increasingly available from Chinese OEMs at modest premium.

USB3 Vision (USB 3.2 Gen 1, 5 Gbit/s) is appropriate for bench-top systems, compact embedded deployments, or where PCIe slots for frame grabbers are unavailable. Cable length is the hard constraint: the USB 3.2 specification allows 3 m for passive cables; active optical cables extend to 10–15 m at additional cost. USB3 Vision cameras are plug-and-play on Windows without driver installation if the host meets USB 3.0 hub requirements, which simplifies embedded system integration (e.g. NVIDIA Jetson, Raspberry Pi CM4, standard x86 mini-PCs).

GenICam standard compliance is more important than the transport layer. GenICam (Generic Interface for Cameras) defines a uniform XML-based camera description file that any compliant software framework reads to enumerate the camera’s feature set — gain, exposure, pixel format, trigger mode, ROI, and so on. When a camera is genuinely GenICam compliant, it works with Halcon (MVTec), VisionPro (Cognex), OpenCV + Aravis (open-source GStreamer/V4L2 stack), National Instruments Vision, and Labview IMAQ without vendor-specific plugins. This is the critical quality differentiator between credible Chinese industrial cameras and cheap cameras marketed as industrial.

The common failure mode with non-compliant Chinese cameras: the GenICam XML file provided with the camera has structural errors (malformed feature descriptions, missing mandatory features, incorrect access mode declarations) that cause SDK load failures. The camera appears in the network scan, then throws an exception during device feature enumeration. Always validate GenICam compliance before committing to volume — request a sample, connect it to your production SDK, and iterate through all feature nodes programmatically before approving the design.

Hardware trigger and jitter. For high-speed lines above 60 frames/s, software trigger over GigE introduces timing jitter in the 1–5 ms range due to OS scheduling latency. An isolated optocoupler GPIO trigger at the camera hardware pin reduces trigger-to-exposure jitter to <1 µs, which is necessary for synchronising camera exposure to encoder pulses on a moving conveyor. Confirm the factory’s camera supports both trigger delay (set in µs via GenICam feature) and trigger overlap mode for continuous high-speed acquisition.

Firmware version stability. A recurring issue with Chinese camera OEMs at lower price points is inconsistent pixel format support across firmware versions — a camera shipped with firmware 1.3.x supporting Mono8 and BayerRG8 may ship in a later production batch with firmware 2.0.x that drops BayerRG8 or changes the GenICam XML structure. Build firmware version pinning into your sourcing agreement and confirm that the factory maintains access to historical firmware images for at least 3 years.

Quality Assessment and Factory Evaluation

EMVA 1288 characterisation. The European Machine Vision Association’s Standard 1288 defines a measurement methodology for camera performance parameters: readout noise, dark current, system gain, absolute sensitivity, dynamic range, and SNR curves. A reputable Chinese industrial camera factory should be able to provide an EMVA 1288 report for each sensor model. If they cannot — or provide a report with measurement conditions undisclosed — treat that as a quality signal. The measurements you care about most: readout noise (<3 e⁻ for Pregius S sensors is achievable), dark current at operating temperature, and full well capacity (which sets the upper bound of dynamic range alongside readout noise).

Pixel defect map. Every CMOS sensor ships with a manufacturer defect specification listing dead pixels (no response to light), stuck pixels (always saturated), and hot pixels (elevated dark current). Request the factory’s incoming inspection process for pixel defects and their acceptance threshold. The appropriate threshold varies by application: industrial microscopy and semiconductor inspection tolerate zero cluster defects; label inspection and dimensional measurement typically allow <50 defect clusters per megapixel. Ask whether the camera applies pixel defect correction in hardware (FPGA-based look-up table interpolation) and whether the correction map is user-updatable.

Thermal stability. Sensor gain and offset drift with temperature. For a factory floor with ambient temperature variation above 20°C (common in unair-conditioned facilities), specify a camera with onboard temperature compensation or a factory-calibrated temperature coefficient file. Cameras without compensation will drift in effective brightness across shifts, introducing false rejects or false accepts as ambient temperature rises through the day. Ask the factory to supply a gain-vs-temperature curve across the operating range (typically 0–50°C or -10–60°C for industrial grade).

Lens mount precision. C-mount cameras have a nominal flange focal distance of 17.526 mm. Manufacturing tolerance on this dimension matters for repeatable focus when cameras are swapped across production fixtures. A tolerance of ±0.01 mm or better ensures that replacing a camera body does not require refocusing the lens. Ask the factory for their CMM measurement records on a sample of housings, not just the nominal value in the datasheet.

Factory evaluation. For a factory audit of a machine vision camera supplier, the assessment should go beyond the standard factory checklist. Insist on a working machine vision demonstration — a live inspection line using the cameras you are evaluating, not a bench demo of a single camera pointed at a static target. This reveals integration quality: how the factory handles trigger timing, illumination control, and image analysis pipeline integration. Factories that can only demo cameras in isolation are rarely equipped to support your integration in production.

For volume requirements above 50 units, a pre-shipment inspection should include sampling for EMVA 1288 parameters against a reference unit to verify production consistency. Camera performance parameters are sensitive to PCB layout, shielding, and sensor bonding quality — batch-to-batch variation is a documented issue with second-tier Chinese suppliers. A dedicated quality inspection protocol for camera optics and electronics will catch problems before they reach your facility.