Sequential exposure time mode: how depth cameras beat mixed scenes

On a depth camera's exposure menu you usually find three entries: a single fixed exposure time, an auto exposure, and one more called sequential exposure time mode. That third entry decides whether a cell survives a scene that mixes black rubber with polished steel — and it also decides your frame rate. This page is for whoever sets exposure on a depth head and has to defend the cycle time afterwards. It works through what the camera captures in that mode, how it fuses the captures into one depth frame, and what the mode costs, with the arithmetic to make the call before you wire the cell.

What sequential exposure time mode does

Select the mode and the camera changes what one measurement means. Instead of a single capture at one exposure time, it fires a burst: the same scene, the same optics, the same geometry, captured several times at increasing exposure times, typically spaced by a constant factor between neighbours. The fusion engine inside the camera merges that burst into a single depth frame — same format, one result per trigger.

The property you buy is coverage. For every pixel, at least one capture in the burst places the returned signal inside the sensor's usable window: bright enough to range, below saturation. Exposure menus label the idea HDR, HDR+ or sequential exposure time mode; the mechanism is the same multi-exposure HDR capture used in photography, applied per pixel to depth. The mode's promise is that a mixed scene no longer comes back half-empty. The bill is cycle time, and the worked example below prices it.

Why one exposure time cannot cover a real scene

Two ends fail a single exposure. Above saturation, an over-exposed pixel degrades before it clips outright: on ToF and structured-light heads, depth accuracy falls away as the charge piles up. Below the noise floor, the pixel reports noise instead of range. Between the ends sits the sensor's usable window.

Signal scales linearly in the two directions that matter: double the surface reflectance and the returned signal doubles; double the exposure time and it doubles again. So the exposure you pick has to pull the whole scene into one window. Black rubber sits near 5% reflectance, machined steel at 85% or more — a 17:1 spread before distance and angle contribute. Returned amplitude also falls roughly with the square of distance, so unequal stand-offs stretch the demand further. Set the exposure for the steel and the rubber returns counts buried in noise; set it for the rubber and the steel saturates. When that spread outruns the window, no single exposure time works — the exact situation the sequential mode exists to remove.

How the camera fuses the burst into one depth frame

Fusion runs per pixel. Across the burst, the camera keeps, for each pixel, the capture whose signal landed in the valid band. Captures where that pixel saturated are discarded even though other pixels in the same capture were fine; captures where the signal never cleared the noise floor are discarded too. What reaches your application is one depth frame plus a validity or confidence map that says which pixels to trust. Read the map before you trust the frame — confidence image values on a 3D camera: decode bit flags and mask bad pixels covers exactly this masking job, and it applies here unchanged.

One assumption sits underneath: the scene holds still for the whole burst. Keep that in mind until the motion numbers below.

What the mode costs: measurement time, frame rate and data

Every capture in the burst pays its own overhead — readout, transfer, processing — on top of its exposure time, so the measurement time adds up:

Here t_meas is the measurement time in ms, N the number of exposure sets, t_i the individual exposure times, and t_overhead the per-capture cost you already pay in single-exposure mode. You can measure t_overhead on your own head: run single-exposure mode, take the frame period, subtract the exposure you set. Treat the formula as a planning estimate — pipelines that overlap part of the overhead come in faster — so time both modes before you freeze the cell design. On ToF heads the coupling runs the other way too: the ToF camera exposure time and frame rate trade-off explains why a longer integration alone already moves the frame rate.

Where it earns its keep, and where motion kills it

Enable it when one field of view spans the reflectivity range: rubber seals beside steel brackets, dark plastics in a glossy tray, dark housings behind glass, bins where stand-off changes from part to part. The arithmetic above flips to the mode's advantage. In the example scene, the 0.25 ms capture holds the steel at 255 × 0.25 ≈ 64 counts — mid-scale instead of saturated — while the 16 ms capture lifts the rubber from 15 counts to 15 × 16 = 240 counts. Both surfaces land inside the usable window and survive into one fused frame.

Leave it off when the scene is matte and uniform and the takt is tight. A single fixed exposure then does the same job in a fraction of the time, with nothing to fuse and nothing to mis-fuse.

Motion is the mode's hard limit. The burst spans tens of milliseconds, so a part that moves during it presents a different scene to each exposure, and fusion meets pixels that are valid in one set and empty in the next — the same artifact class as on-sensor HDR versus multi-exposure motion artifacts, showing up here as ghost edges and torn validity flags. The budget is unforgiving: on a conveyor at 300 mm/s, a 75 ms burst moves the part 300 × 0.075 = 22.5 mm. Stop the part, or run one short exposure.

How the exposure modes compare

Auto exposure has its place during commissioning: set a brightness target that matches the inspection, then lock everything in production, because auto modes are the first suspect when image brightness drifts over time. For 2D work, the HDR capture and fusion workflow for surface inspection and the multi-exposure settings consistency guide cover the host-side version of the same trade.

Checks before you enable the mode

• Measure both frame periods. Single-exposure mode gives you the per-capture overhead; sequential mode gives you the real cycle. The worked example's gap, 50 fps down to 13 fps, is the size of change to expect.
• Budget the takt with the formula. Measurement plus transfer plus processing must fit the cycle; 75 ms of measurement already consumes most of a 100 ms takt.
• Freeze the part for the burst, or prove its travel during the burst stays well inside your lateral tolerance — the 22.5 mm conveyor case above is the scale of the problem.
• Read the confidence map after enabling. Saturation flags on shiny faces should clear, and the dark surfaces should gain valid depth.
• Set the ladder deliberately where the menu exposes the times, using the reflectance-ratio tip above; otherwise record what the camera picked so you can reproduce it.
• Re-check link load if you stream raw sets: per-frame raw volume scales with the number of sets, and the software-trigger frame-rate isolation walkthrough separates capture cost from stream cost fastest.
• Lock the mode in production. No auto exposure anywhere in the chain; drift starts there.

Depth cameras that give you the choice

The 3D sensors range spans the head types where this decision shows up: ToF, projected structured light and laser scanning. The camera finder filters them by type and range, and the configurator walks the rest of the stack around the head you pick.

The MV-DT01WMU is a ToF sensing camera for volume and presence work — the head type whose shiny-surface dropouts this mode is built to fix. The MV-DPS200P-02 is a projected structured-light 3D camera, where the exposure ladder trades against the projector's own brightness. The MV-DLS300P-04 pairs a laser galvanometer with stereo capture, so scan-line alignment works alongside exposure as a lever.

More depth-capture guides sit on the 3D vision topic page. To have this trade-off priced for your scene — surfaces, reflectivity spread, stand-off, cycle target — send those four numbers through the contact page, and we will come back with a mode recommendation and a measurement-time estimate for the specific head.