A robot vision system is a camera, a lens, a light, a mount and software — not a camera alone. Teams that buy the camera first and discover the rest later usually spend twice. The camera is often the cheapest part of a working inspection or guidance setup.
This guide covers the sensor choices that matter for robotics, the interfaces you will meet, what industrial vision hardware costs in India, and how to size a camera to the job before you order anything.
Start with the task, not the megapixels
Every vision task resolves to two numbers: the field of view you must cover, and the smallest feature you must detect. Divide the field of view by the feature size to get the minimum number of pixels across the image. Then add margin, because you rarely get perfect lighting or perfect alignment.
For example, inspecting a 300 mm wide tray for a 1 mm defect needs at least 300 pixels across, and realistically 600–900 to detect it reliably. A 1280-pixel-wide sensor is comfortable. A 5-megapixel camera on the same job is not “better”; it is more data to process, more light needed and a narrower choice of lenses.
Rolling shutter versus global shutter
| Property | Rolling shutter | Global shutter |
|---|---|---|
| Exposure | Rows exposed sequentially | Whole sensor exposed at once |
| Moving objects | Skew and wobble | Clean geometry |
| Cost | Lower | Higher |
| Typical use | Static inspection, consumer cameras | Conveyor tracking, robot-mounted cameras |
If the camera or the part moves during exposure, use a global shutter. This is the single most common specification mistake in robot vision projects: a cheap rolling-shutter camera works beautifully on a bench and produces unusable images once mounted on a moving arm.
Interfaces and cabling
- USB 3.0. Simple, widely supported, cable length limited to a few metres. Good for bench and small cells.
- GigE Vision. Long cable runs over standard Ethernet, power over Ethernet in many models, multi-camera friendly. The default for industrial cells.
- MIPI CSI. The ribbon-cable interface on Raspberry Pi and Jetson boards. Cheap and fast, but short cables and board-specific drivers.
- Camera Link. High bandwidth for line-scan and very high-resolution cameras; needs a frame grabber.
For a Raspberry Pi or Jetson project, a CSI camera is the pragmatic choice. For a production cell with four cameras and 15 m cable runs, GigE with a managed switch is easier to maintain.
Lenses and lighting matter more than the sensor
A camera sees contrast, not objects. If the defect has the same brightness as the background, no amount of resolution will find it. Lighting is what creates contrast, and it is the part of the system most often left to chance.
- Backlight. Silhouettes for measuring outlines, holes and presence.
- Ring light. Even illumination for flat surfaces and general inspection.
- Bar or dome light. Controlled reflections on shiny, curved or machined parts.
- Coaxial light. On-axis illumination for flat reflective surfaces such as wafers and glass.
- Structured light or laser. Height and profile, not just colour.
Lens choice follows the geometry. A short focal length gives a wide view from close range; a long focal length gives a narrow view from far away. Fixed-focal-length lenses are sharper and cheaper than zoom lenses, and industrial mounts (C-mount, CS-mount) are not interchangeable with consumer camera lenses.
What vision hardware costs in India
Indicative 2026 planning ranges, before GST. Industrial pricing varies with resolution, frame rate, interface and brand.
| Item | Indicative range | Notes |
|---|---|---|
| Raspberry Pi camera module | ₹1,500–6,000 | CSI, rolling shutter, good for learning |
| USB webcam, 1080p | ₹2,000–8,000 | Rolling shutter, poor low light, fine for demos |
| USB3 industrial camera, 2–5 MP | ₹18,000–60,000 | Global shutter models cost more |
| GigE industrial camera, 2–5 MP | ₹30,000–90,000 | Long cable runs, PoE options |
| C-mount lens | ₹4,000–30,000 | Fixed focal length; aperture and distortion matter |
| Ring or bar light with controller | ₹3,000–25,000 | Strobing controllers cost more |
| Smart camera with built-in tools | ₹60,000–2,50,000 | Vendor software included; less flexible |
| Depth camera | ₹20,000–90,000 | RealSense, Orbbec and similar |
A workable single-camera inspection setup with a global-shutter industrial camera, lens, light and mount typically lands between ₹50,000 and ₹1.5 lakh before software and integration. A demonstration setup using a Pi camera can be built for under ₹10,000.
Software choices
OpenCV is the default for custom work: it runs on Pi, Jetson and industrial PCs, and covers calibration, thresholding, contour detection, template matching and camera streaming. Vendor smart-camera tools are faster to deploy for standard inspection, but lock you into their hardware and licensing.
For robot guidance, the software must also output a pose in the robot’s coordinate frame, which means hand-eye calibration. Budget time for that calibration; it is the step that turns a detection into a usable pick coordinate.
Mistakes that cost the most
- Buying resolution instead of contrast. Fix lighting first; a 2 MP image with good light beats a 20 MP image with glare.
- Ignoring motion. Rolling shutter on a moving robot produces distorted geometry that no algorithm can fix.
- Forgetting calibration. A camera that sees the part but does not know where the robot is cannot guide it.
- Under-specifying the mount. A camera that vibrates by 0.1 mm at 30 fps is measuring vibration, not the part.
- No plan for ambient light. Factory windows and skylights change illumination through the day; enclose or control the scene.
- No trigger strategy. Software triggering is fine for slow lines; hardware triggering is safer for fast ones.
Related: OpenCV on Raspberry Pi, Raspberry Pi 5 robot guide, distance sensor comparison, delta robots, SLAM for beginners.
What camera is best for robot vision?
A global-shutter industrial camera with a suitable lens and controlled lighting is best for moving robots and conveyors. For learning, a Raspberry Pi camera or USB webcam is enough to practise OpenCV.
How much does a machine vision camera cost in India?
Indicative 2026 ranges: USB3 industrial cameras ₹18,000–60,000, GigE cameras ₹30,000–90,000, C-mount lenses ₹4,000–30,000, and lighting ₹3,000–25,000, before GST.
Do I need a global shutter camera?
If the camera or the part moves during exposure, yes. Rolling-shutter sensors distort moving objects, which breaks measurement and pose estimation.
How many megapixels do I need?
Divide the field of view by the smallest feature you must detect, then add margin. A 2–5 MP camera covers most robot guidance and inspection tasks.
Can I use a webcam for robot vision?
For a demo or a slow indoor robot, yes. For production, its rolling shutter, auto-exposure and plastic lens mount make results unpredictable.
Vision-guided picking needs both hand-eye and TCP calibration.
Camera choice and compute choice are linked; see Raspberry Pi vs Jetson for robotics for the processing side.
Camera realism affects perception testing: Isaac Sim vs Gazebo compares rendering fidelity.
For the lowest-cost option, see ESP32-CAM modules and their limits.
The lighting half of that chain is in the machine vision lighting guide.
A camera alone does not make a cell pick reliably, so plan the transform with the vision-guided robotics guide.
Lens choice sets the field of view, but camera calibration with OpenCV is what turns that view into measurable millimetres.
If the job needs distance rather than a flat image, compare the options in the depth camera buying and setup guide.
Camera choice follows from the defect size, which this inspection guide works through with arithmetic.
Once the camera type is shortlisted, check the interface against your cable run and frame rate with our industrial camera interface comparison.
The camera body is half the optical system, and the lens selection guide covers focal length, sensor coverage and cost bands.
Bin picking is the hardest case for a camera, so choose the sensor with depth quality and cycle time in mind.
Sources and further reading
- EMVA 1288 — standard for camera sensitivity, noise and dynamic range measurements.
- Basler — industrial camera interfaces, sensor comparisons and lens calculators.
- OpenCV — camera calibration, detection and image processing documentation.
- Teledyne FLIR — machine vision cameras and lighting guidance.
