Bright light in a depth camera’s view can produce apparent obstacles even when the physical scene has not changed. The immediate challenge is separating an optical measurement error from a real object, a coordinate error, or stale data retained downstream. This troubleshooting sequence helps you compare camera aim, temporary optical shielding, and exposure settings while checking that genuine obstacles remain measurable. Make changes in a controlled test area, with machine motion stopped or otherwise managed by the established safety procedure.
Separate false depth from missing or misplaced data
Start by comparing the camera’s native depth image with the point cloud and the final obstacle map. Save synchronized captures where the software permits it. A false obstacle can originate at any of these stages, and the correction depends on where it first appears.
Look for these distinctions:
- Spurious valid depth: pixels contain plausible distances where the scene should be empty or substantially farther away.
- Invalid depth: the camera reports missing measurements, but downstream software interprets them as nearby surfaces.
- Misplaced geometry: the depth image looks reasonable, but a coordinate transform puts points in the wrong location.
- Retained geometry: the current point cloud is clear, but an obstacle map still contains older observations.
Check the camera documentation for invalid-depth encoding. Zero, a maximum value, and a non-finite value are not interchangeable conventions. Invalid measurements mean unknown space unless another validated measurement establishes otherwise.
If the main symptom is absent geometry rather than invented surfaces, follow the separate guide to missing points in a camera point cloud. If geometry appears rotated or displaced, investigate the optical-frame and camera-frame transform before changing exposure.
Test the likely causes in order
When the symptom follows a bright source entering the view, investigate optical interference first. However, brightness alone does not identify the mechanism, and different depth technologies fail differently.
| Likely cause | Evidence to look for | Controlled test | Practical correction |
|---|---|---|---|
| Saturation or flare | Bright or washed-out regions coincide with bad depth | Move the source outside the view without removing the test target | Change aim, source position, or optical shielding |
| Exposure compromise | Bright regions improve while darker targets lose depth | Compare fixed exposures with the same scene | Select an operating exposure that preserves required targets |
| Stereo matching failure | Incorrect depth follows lost texture or unequal views | Inspect both input images where available | Restore usable image detail and remove direct glare |
| ToF interference or indirect light paths | Depth changes near bright sources or reflective surfaces | Change source visibility and reflective geometry separately | Reduce direct illumination or problematic reflected paths |
| Processing or map persistence | Raw depth clears before the obstacle map does | Replay or inspect synchronized processing stages | Correct invalid-data handling, transforms, or clearing logic |
A normal-looking color image does not rule out an optical problem in the depth channel. Its wavelength response, viewpoint, and exposure may differ from the depth receiver.
The related article on the IR bright-light autoexposure floor addresses exposure limits and noisy measurements. Here, the key question is whether source visibility causes false occupied space and whether the proposed correction preserves the inspection scene.
Establish a repeatable baseline
Choose a fixed test region containing both an empty background area and the positions where real targets must be detected. Record the required near and far limits, smallest relevant target, and acceptable detection latency before adjusting the camera.
- Save the initial configuration. Record camera pose, exposure mode, exposure setting, gain, depth-processing options, and map settings.
- Capture an empty-scene run. Count frames containing a false obstacle inside the relevant region, not across the whole image.
- Capture a target-present run. Place an inert target at known positions and record whether its depth is usable.
- Change source visibility only. If permitted, switch or reposition the troublesome light without changing the target or camera settings.
- Repeat the original condition. An A/B/A sequence helps distinguish a repeatable optical effect from drift or a transient software condition.
Use a temporary screen between the light and camera only when it stays outside the required sensing region. Blocking the camera’s entire view is not a valid obstacle-detection test.
Repeat after exposure has settled and after a fresh startup. If the scene changes mainly during initialization, use the guide to depth coverage changes after startup to separate settling behavior from persistent interference.
Change camera aim and shielding first
There is no universal mounting-angle limit that guarantees reliable depth. The relevant factors include source position, receiver coverage, projector coverage where applicable, surface reflectivity, and the scene you must measure.
Try small, documented aim changes rather than moving the camera until the image merely looks cleaner. For example, compare the original pose with a 5-degree change, then verify coverage before trying another increment. These are experimental increments, not camera specifications or recommended limits.
Check the entire required working volume after every change. A target visible in the color image may sit outside valid depth coverage, particularly where stereo views overlap only partially. The guide to why depth coverage can be smaller than image coverage explains that distinction.
A temporary matte optical hood can help test off-axis stray light. Keep it clear of every receiver, projector, and other required optical opening. Check the image edges and the nearest required target for clipping. A hood cannot block a source that occupies the same viewing direction as a required target without also obstructing that target.
For wider placement decisions, compare the required scene against the 3D camera working-distance and positioning guide. Do not exchange a glare problem for an uncovered approach path.
Test exposure without hiding real obstacles
Where supported, temporarily select fixed exposure to make optical comparisons repeatable. Keep gain and processing settings unchanged during that comparison. Restore automatic operation later only if the application needs it and its full operating range has been verified.
As an illustrative exposure bracket, compare 4 ms, 2 ms, and 1 ms if the camera supports those settings. Reducing exposure from 4 ms to 1 ms reduces integration time to one quarter. Under otherwise unchanged, unsaturated conditions, the accumulated light signal also falls approximately to one quarter.
That reduction affects useful light as well as unwanted light. Dark targets, distant surfaces, and projected patterns can become harder to measure. Therefore, compare target coverage and depth stability alongside the false-obstacle count.
For stereo systems, examine both source images where the interface permits it. For ToF systems, inspect amplitude, confidence, or validity information if those outputs are documented. Do not assume that a similarly named confidence value means the same thing across technologies.
Repeat tests with the relevant lighting states and source positions. Lighting modulation can make one short recording unrepresentative. Avoid selecting a setting from a single favorable frame.
Compare false detections with retained coverage
Use two separate measurements: false-obstacle frames in an empty scene and successful detections with a real target present. Combining them into one score can hide a dangerous tradeoff.
Consider this hypothetical commissioning example. Each configuration has an empty-scene recording and a separate target-present recording, each containing 1,000 frames. At a recorded rate of 25 frames/s, each recording lasts:
| Configuration | Empty-scene frames with false obstacles | Target-present frames with usable detection |
|---|---|---|
| Original setup | 240 | 995 |
| Aim changed | 8 | 996 |
| Temporary optical hood | 0 | 996 |
| Shorter exposure only | 0 | 810 |
| Aim change and hood | 0 | 997 |
These numbers illustrate a decision method, not measured product performance. Shorter exposure appears successful if you inspect only false obstacles, but it substantially reduces target detection. The geometry changes are better candidates for further testing in this example.
Zero observed false detections does not establish a zero failure rate. Consecutive frames may also be strongly correlated. Extend testing across target positions, lighting conditions, startup cycles, and relevant motion before accepting a configuration.
Verify the downstream processing separately
Once the native depth measurement improves, trace the correction through the pipeline. Confirm that timestamps, coordinate transforms, point-cloud generation, and obstacle-map clearing remain consistent.
A region of interest can exclude a light fixture that lies outside the application’s required volume. It is not an optical repair: flare originating outside the retained region can still affect measurements inside it. Spatial cropping must also leave every required target path covered.
Temporal filtering can reject isolated points, but it adds delay and may remove brief observations of small or moving targets. For example, requiring three consecutive detections at 30 frames/s spans about 67 ms between the first and third observations. Exposure, transfer, and processing add further latency.
Document that delay against the application’s response requirement. Use the related guide to depth-map change detection without speckle false alarms when designing persistence rules, rather than increasing filtering until the display looks quiet.
Never convert invalid pixels into confirmed free space merely to clear an obstacle map. Preserve an explicit unknown state where the measurement does not support an occupancy decision.
When changing the sensor is the right next step
If the bright source must remain in the required view, a hood may not solve the problem. If reduced exposure destroys target coverage, software adjustments may have reached a practical limit. At that point, compare sensing technologies using representative scene trials.
The catalog identifies MV-DT01WMU as a ToF sensing camera. It is a candidate for a technology-specific evaluation, not evidence that ToF will reject the troublesome light.
- Type
- Depth / ToF camera
The catalog identifies MV-DPS200P-02 as a projected structured-light camera. Its suitability likewise requires testing against the actual illumination and target surfaces.
- Type
- Stereo / structured light
Browse 3D sensors or the camera finder to identify candidates by sensing method. Request the manufacturer’s documented ambient-light conditions, operating range, output-validity conventions, and supported exposure controls before selecting a replacement. A different technology changes the failure mechanisms; it does not eliminate the need for validation.
Checks before returning the system to service
Use this final checklist to prevent a visually convincing fix from becoming a coverage problem:
- Empty-scene behavior: repeat the source-visible and source-hidden conditions without changing processing.
- Required targets: test representative sizes, finishes, orientations, and distances throughout the operating volume.
- Coverage boundaries: inspect image edges, overlap limits, and any area shadowed by a hood.
- Transitions: repeat startup, exposure settling, and relevant lighting changes.
- Timing: measure detection and clearing delays through the final application output.
- Unknown data: confirm that missing measurements remain distinguishable from clear space.
- Configuration recovery: save settings and mounting references, then verify them after a restart.
- Safety separation: retain the machine’s established protective functions and complete the required validation before operation resumes.
If geometry changes resolve the problem, document them before applying additional filtering. If no setting preserves the required targets, bring representative depth captures, exposure records, and scene dimensions to application support before ordering a different sensor.
Questions
Is there a maximum camera tilt that prevents bright-light errors?
There is no mounting angle that is reliable for every depth camera and scene. Evaluate whether the troublesome source enters the receiver view, whether required targets remain covered, and whether the depth method still works on their surfaces. Use the manufacturer’s placement guidance and verify the complete operating volume after each adjustment.
Will reducing exposure remove false obstacles?
It can reduce saturation, but it also reduces the useful signal from targets or projected illumination. Accept the change only after testing representative dark, distant, small, and angled targets. A lower false-obstacle count is not sufficient if genuine objects become invalid or disappear.
Can an optical hood help when the light is directly in view?
A hood mainly helps with unwanted light arriving outside the required viewing directions. If the source and a required target share the same direction, blocking one can block the other. Test alternative camera placement, source placement, or another sensing approach instead of extending the hood into the measurement region.
Should point-cloud filtering be the first correction?
Filtering is useful for characterized outliers, but persistence and neighborhood rules can also reject small or briefly visible targets. Establish the raw measurement problem first, then quantify added latency and retained detection coverage. Keep invalid measurements separate from confirmed empty space.
When should a different depth sensor be evaluated?
Consider another sensor when the interfering source cannot be excluded and supported settings cannot preserve the required scene. Compare candidates using the same lighting, targets, distances, and acceptance criteria. Catalog sensing method alone does not establish ambient-light tolerance or application performance.