Depth image field of view: why stereo coverage is smaller

A raw infrared image can show a surface that never appears in the depth map. That does not necessarily mean the sensor is partly disabled. Stereo depth requires usable observations from both cameras, followed by rectification, correspondence matching and output formatting. Each stage can reduce the region containing trustworthy measurements.

For inspection and positioning tasks, compare the usable depth footprint at your actual working distance, not just the image dimensions. The practical task is to separate geometric coverage, output cropping and failed measurements before choosing a camera or operating mode.

Separate four different kinds of coverage

The depth image field of view is not necessarily the field of view of either individual camera. It can describe several different boundaries, depending on the documentation or software display.

An image can retain its full rectangular dimensions while containing invalid depth along one edge. Conversely, a smaller output can contain valid measurements across nearly its entire width. These are different problems and need different remedies.

Treat pixel dimensions as the size of the delivered array, not proof of an angular or physical footprint. The guide to live stereo rectification explains how paired images are transformed before correspondence matching.

Shared coverage depends on working distance

For a simplified stereo pair, assume parallel optical axes, equal focal lengths and a horizontal baseline. A surface must fall inside both viewing regions before stereo correspondence is possible. Near the cameras, their viewing regions overlap less than they do farther away.

The field-of-view diagram and accompanying angle calculation explain the single-camera geometry. The figure below illustrates that individual viewing cone; stereo adds a second, displaced cone whose intersection matters.

For an ideal rectified pair viewing a plane perpendicular to the optical axes:

Here, Z is axial distance from the camera plane, Nx is image width in pixels, fx is horizontal focal length in pixels, and B is baseline. Use the same length unit for distance, baseline and footprint width. These equations are geometric estimates, not guaranteed coverage specifications.

Consider an illustrative pair with an image width of 1280 pixels, a focal length of 800 pixels and a baseline of 60 mm. At an axial distance of 600 mm:

Each camera sees a footprint 960 mm wide, but their common footprint is only 900 mm wide. On this plane, the corresponding horizontal image shift is 80 pixels.

At 300 mm, the same ideal geometry gives an individual footprint of 480 mm and shared coverage of 420 mm. The missing fraction therefore grows at closer range. These calculations exclude distortion, rectification crops, occlusion and matching support requirements.

Baseline also affects depth sensitivity, so reducing it is not automatically the right solution. Read wide baseline stereo camera selection when balancing coverage against depth measurement requirements.

Output aspect ratio can remove a different part of the scene

Horizontal overlap does not, by itself, explain every vertical crop. Output modes can use different sensor regions, rectification maps, resizing operations or aspect ratios. A height reduction may therefore come from formatting rather than the physical baseline.

Suppose a hypothetical raw image contains 1280 × 800 pixels. A depth output contains 1280 × 720 pixels. If the depth mode preserves the same pixel scale and uses a centered vertical crop, it retains:

That would remove 10% of the raw vertical coverage, split equally between top and bottom for a centered crop. In the earlier example, an ideal vertical footprint of 600 mm would become 540 mm.

However, those conclusions apply only if the operation is actually a crop. A resized image could preserve the field of view while reducing sampling density. A different readout mode could change both coverage and sampling.

Check the mode documentation and calibration parameters rather than inferring the operation from resolution alone. The explanation of undistortion scaling and field of view helps distinguish retained scene coverage from output image size.

Invalid disparity borders are not necessarily cropped pixels

A stereo matcher compares a reference-image location with candidate locations in the other image. Near an edge, the required corresponding location may lie outside the other camera's image.

In the worked example, a point on the target plane has a horizontal disparity of 80 pixels. Some edge locations in the reference image therefore lack a visible counterpart. Which edge is affected depends on the reference camera and disparity convention; do not assume identical missing strips on both sides.

Matching can remove additional coverage because it needs a neighborhood around each candidate. Some implementations also reserve borders for the configured disparity search range. Others evaluate candidates near the boundary differently. The exact behavior is implementation-specific.

Even inside the geometric overlap, valid depth is not guaranteed. Occlusion, low texture, reflections, saturation and confidence filtering can leave holes. A foreground edge can hide a background surface from one camera while leaving it visible to the other.

If the boundary is irregular or follows object edges, investigate correspondence quality before changing output resolution. Use stereo disparity map troubleshooting to separate alignment problems from matching failures.

Verify usable coverage with a controlled scene

Use a flat, textured target larger than the required inspection region. Keep it stationary and place it at the nearest, nominal and farthest intended working distances. A single test at the nominal position can conceal a close-range limitation.

1. Record the mode configuration. Save stream dimensions, calibration selection, rectification settings, depth range and filtering options. Confirm which camera supplies the reference image.
2. Capture both raw views. Identify target features near every edge. A feature missing from either raw image cannot produce ordinary stereo correspondence.
3. Inspect the rectified pair. Check whether those same features survive the remapping. A rectangular missing strip introduced here points toward the rectification or output definition.
4. Inspect disparity and validity. Separate invalid pixels from legitimate near or far measurements using the software's documented validity representation.
5. Measure the usable footprint. Record physical width and height, the valid-pixel fraction, and the largest uninterrupted valid region covering the task.
6. Repeat with representative parts. A textured flat target establishes geometric potential, but reflective, dark or deeply recessed parts may deliver less usable coverage.

Define acceptance around the actual inspection region. For example, a task may need valid measurements throughout a central 500 mm × 300 mm rectangle, rather than a high average validity score across the whole image.

The guidance on calibration target coverage margins helps size the target and leave room for positioning variation. If empty corners appear only after remapping, compare the symptoms with black borders after undistortion.

Budget usable pixels as well as physical width

A footprint can be large enough yet too sparsely sampled for the smallest feature. Calculate both requirements before accepting a mode.

Suppose the required inspection width is 600 mm. The smallest feature is 3 mm wide, and the task uses a planning target of 5 pixels across that feature:

The 5-pixel target is an application assumption, not a universal detection rule. Contrast, optics, matching quality and the measurement algorithm determine whether it is adequate.

Count usable pixels over the required footprint, not every column in the delivered array. Upsampling a depth map increases its dimensions but does not create additional independent measurements. Likewise, a confidence mask can reduce effective coverage without changing the nominal resolution.

Spatial sampling also differs from depth precision. Enough pixels across a feature does not establish acceptable uncertainty along the viewing direction. Validate that requirement separately with representative surfaces and distances.

Compare operating modes before comparing products

Start a mode comparison with a coverage worksheet. For each candidate, record raw-view coverage, rectified coverage, valid depth footprint, acquisition timing and measurement stability at the same distances.

A larger output mode is useful only when it improves the requirement that currently limits the application. If it merely interpolates an existing result, it may add processing load without improving coverage or detail.

The 3D sensors category and camera finder provide starting points for comparing acquisition architectures. Do not transfer the passive stereo assumptions above to every active depth system without checking its documented operation.

The MV-DLS300P-04 is listed as a laser galvanometer stereo camera.

The MV-DPS200P-02 is listed as a projected structured-light camera.

These catalog descriptions identify acquisition approaches; they do not establish a particular depth footprint or output crop. Request mode-specific coverage information and representative captures before selecting either for a coverage-critical task. The illustrative calculations above are not specifications for these products.

Check these items before ordering

Use a short acceptance checklist that distinguishes the scene you need from the image the software displays:

• Working-distance envelope: include the nearest surface, not just the mounting distance to the conveyor or fixture.
• Required footprint: specify physical width and height, plus placement tolerances and motion margins.
• Depth reference: identify the coordinate frame and image to which depth values are registered.
• Output mode: confirm whether differences come from cropping, resizing, sensor readout or rectification.
• Usable sampling: state the minimum valid region and required feature sampling, not resolution alone.
• Surface behavior: test texture, gloss, dark materials and recesses that can cause missing correspondence.
• Timing conditions: use the intended exposure, motion and acquisition sequence during acceptance testing.

Ask for measurements at the operating distances rather than one advertised viewing angle. A single angular figure cannot describe every near-range overlap boundary or scene-dependent validity limit.

Know when the usual explanation is wrong

Not every reduced-looking depth image is a stereo coverage problem. A viewer may stretch images into different windows, apply digital zoom or display only a selected region. Compare saved frames and metadata before drawing conclusions from screen appearance.

Registration into another camera's coordinate frame can also change the visible footprint and create holes. Resolution changes require the corresponding calibration information; reusing incompatible intrinsics can produce misleading geometry rather than a genuine field-of-view reduction.

If the depth boundary changes with illumination or surface texture while the raw scene remains unchanged, focus on measurement validity. The discussion of stereo depth lighting for dark surfaces addresses that distinction.

The right remedy follows the cause: choose another mode for an avoidable crop, revise geometry for insufficient overlap, or improve acquisition conditions for failed matching. Start the camera configuration process, or request help reviewing depth coverage with your required footprint, distance range and paired sample images.