Multipeak Detection in Laser Triangulation: Capturing Multiple Reflection Peaks for Higher Precision

A laser diode emits a focused line onto the measurement object. The object scatters and reflects the laser light back toward the sensor’s receiving optics, which project the reflected line onto a 2D image sensor — typically a CCD or CMOS array. For each pixel column of this array, the sensor computes the intensity distribution along the column axis and identifies the position of maximum intensity. This position corresponds to the z-distance value of the surface at that x-position. The algorithm fits a Gaussian curve to the intensity profile and extracts the sub-pixel centroid as the distance measurement, producing a single z-value per pixel column.

The centroid position z^z^ of a detected peak is calculated by weighted sub-pixel fitting of the intensity values $I(p)$ across pixels $p$ within a fitting window of half-width $w$ around the local maximum at pixel p0p0​:

z^=∑p=p0−wp0+wp⋅I(p)∑p=p0−wp0+wI(p)z^=∑p=p0​−wp0​+w​I(p)∑p=p0​−wp0​+w​p⋅I(p)​

This is the standard single-peak evaluation: one fitting pass per pixel column, one z-value as output.

Multiple peaks arise in 4 physically distinct situations within industrial laser triangulation measurements. First, a transparent or translucent layer — such as a glass pane, polymer film, or protective coating — partially transmits and partially reflects the incident laser beam at each of its 2 interfaces: the front surface generates one reflected peak and the back surface generates a second reflected peak, with both peaks landing at different row positions within the same pixel column. Second, specular or highly reflective surfaces — such as polished metals or lacquered components — generate a direct reflection peak at the correct z-position and one or more interreflection peaks at incorrect z-positions caused by the laser light reflecting between adjacent geometric features before reaching the image sensor. Third, multi-layer composite materials — including adhesive bonds, laminate stacks, and coated substrates — generate one partial reflection peak at each material interface within the laser’s penetration depth. Fourth, a sharp geometry transition — such as an edge, groove, or step height — places 2 surface levels simultaneously within the projection cone of a single pixel column during the transition zone, producing 2 distinct peaks at the z-positions of the upper and lower surface level.

Each detected peak is characterized by 3 measurable attributes. The centroid position, measured in pixels and converted to micrometers or millimeters via the sensor calibration, defines the z-distance value that the peak represents — this is the primary measurement output. The peak amplitude defines the maximum intensity value at the centroid, measured in digital grey-level units on a scale of 0–255 for 8-bit sensors or 0–4095 for 12-bit sensors; amplitude reflects the reflectivity of the surface or layer interface that generated the peak. The peak width, defined as the full-width at half-maximum (FWHM) in pixel units, is determined by surface texture, laser focus quality, and the angle of incidence at the reflecting interface. These 3 attributes — centroid, amplitude, and FWHM — form the basis of all downstream peak selection and validity evaluation decisions.

The FWHM of a Gaussian-fitted peak with standard deviation $\sigma$ is:

FWHM=22ln⁡2σ≈2.355σFWHM=22ln2​σ≈2.355σ

A measured FWHM below the minimum threshold indicates a pixel defect or single-pixel spike; a FWHM above the maximum threshold indicates unfocused volume scatter — both are rejected as invalid peaks.

Transparent and translucent materials — such as float glass, borosilicate glass, polycarbonate films, PMMA sheets, PET packaging foils, and optical adhesive coatings — generate 2 reflection peaks per pixel column: one peak from the front surface and one from the back surface. The front-surface peak has higher amplitude when the surface carries an anti-reflection coating; the back-surface peak has higher amplitude on uncoated glass due to total internal reflection effects at shallow angles of incidence. Both peaks carry geometrically relevant information: the centroid of the first peak defines the front-surface z-position, and the centroid of the second peak defines the back-surface z-position.

The peak separation $\Delta z$ in the sensor coordinate system relates to the physical layer thickness $d$and the material refractive index $n$ through the optical path length:

Δz=2ndcos⁡θΔz=cosθ2nd​

where $\theta$ is the angle of incidence of the laser beam at the layer surface. This relationship is covered in the dedicated layer-thickness metrology article. Single-peak evaluation discards the second peak and reports an incorrect averaged z-value that corresponds to neither surface.

Specular surfaces — including polished steel, mirror-finished aluminum, chrome-plated components, and high-gloss lacquered automotive body panels — produce a direct reflection peak at the correct z-position and one or more interreflection peaks at incorrect z-positions caused by the laser light reflecting between adjacent geometric features before reaching the image sensor. These interreflection peaks are termed ghost peaks. A ghost peak passes the minimum amplitude threshold and occupies a valid pixel column position, causing single-peak evaluation to report the ghost-peak z-position instead of the true surface z-position. Multipeak detection resolves this failure mode by making all peaks visible and applying a peak-ranking rule — typically maximum amplitude selection — to identify the direct reflection peak as the primary output.

Multi-layer composite structures — including glass-fiber reinforced plastics (GFRP), carbon-fiber reinforced plastics (CFRP), adhesive-bonded metal–polymer sandwiches, thermal spray coatings on metal substrates, and printed circuit board (PCB) solder mask layers — generate one partial reflection peak at each internal interface that the laser beam penetrates. The number of detectable peaks depends on the optical transparency of each layer at the laser wavelength — typically 405 nm, 520 nm, or 660 nm for industrial profile sensors — and the layer thickness relative to the sensor’s depth resolution. For a 2-layer bonded system with optically distinct materials, the sensor detects 3 peaks: one at the top surface, one at the bond interface, and one at the bottom surface. Multipeak detection enables profiling of individual layer boundaries that are invisible to single-peak evaluation.

Structured and stepped surfaces — including machined step features, pressed sheet metal profiles, grooves in injection-molded plastic parts, and solder joints on PCBs — produce a transition zone at each geometry discontinuity where 2 surface levels are simultaneously illuminated within the projection cone of a single pixel column. At this transition zone, the image sensor records 2 separate intensity peaks at the z-positions of the upper and lower surface level. Single-peak evaluation averages or interpolates between these 2 peaks, producing a z-value that corresponds to neither surface and falsifying the measured step height. Multipeak detection resolves both peaks independently, enabling accurate step-height measurement down to the sensor’s z-resolution limit.

Scenario Overview

Peak detection identifies candidate peaks within the intensity distribution of each pixel column through a 3-step procedure. In step 1, a global or adaptive intensity threshold is applied to the column intensity vector, retaining only pixels that exceed the threshold value; this step eliminates background noise and sensor dark current. In step 2, local maxima are identified within the thresholded intensity vector: a pixel qualifies as a local maximum candidate if its intensity value exceeds the intensity of its 2 neighboring pixels on both sides. In step 3, a Gaussian curve is fitted to the intensity values in a window of ±3 to ±5 pixels around each local maximum candidate; the sub-pixel centroid of the Gaussian fit defines the z-position of the candidate peak with sub-pixel accuracy.

The centroid resolution $\delta z$ achievable through Gaussian fitting scales with the peak amplitude $A$and the noise floor σnσn​ according to:

δz≈σfitA/σnδz≈A/σn​​σfit​​

where σfitσfit​ is the standard deviation of the fitted Gaussian in pixel units. In practice, centroid resolutions of 1/10 to 1/100 pixel are achieved depending on sensor configuration and surface reflectivity. A minimum peak-separation criterion — defining the minimum distance in pixels between 2 candidate centroids — prevents adjacent noise fluctuations from being counted as 2 separate peaks.

A multipeak-capable sensor applies one of 4 peak-ranking rules to select which detected peak is reported as the primary profile output, or reports all detected peaks as separate z-values. The 4 rules are:

A valid peak satisfies 3 simultaneous validity criteria before being included in the profile output. First, the peak amplitude must exceed a user-configurable minimum amplitude threshold, measured in grey-level units, rejecting peaks caused by laser speckle noise, ambient light, or sensor readout noise. Second, the peak width (FWHM), measured in pixel units, must fall within a user-configurable range defined by a minimum and maximum width value, rejecting single-pixel spikes caused by dust or pixel defects (too narrow) and broad, unfocused scattering contributions from subsurface volume scatter (too wide). Third, the signal-to-noise ratio (SNR) of each peak — calculated as the ratio of peak amplitude to the local noise floor in the surrounding pixel region:

SNR=ApeakσnoiseSNR=σnoise​Apeak​​

— must exceed a minimum SNR value. Peaks that fail any of these 3 criteria are flagged as invalid and excluded from the profile output; the sensor reports a null value or a validity flag at the corresponding x-position.

A multipeak-capable sensor provides 3 configurable output modes. In primary-peak-only mode, the sensor outputs one z-value per pixel column — the z-value of the peak that satisfies the active peak-ranking rule — and the output data format is identical to that of a standard single-peak sensor, enabling drop-in compatibility with existing profile processing software. In all-peaks mode, the sensor outputs up to N z-values per pixel column, where N is the configured maximum peak count per column (typically 2, 3, or 4); each peak is reported with its centroid z-value, amplitude, and validity flag, multiplying the profile data volume by a factor of N. In peak-index mode, the sensor outputs the z-value and amplitude of the peak at a specified rank index (1st, 2nd, or 3rd), enabling systematic extraction of a specific layer boundary from a known multi-layer system without transmitting the full multi-peak dataset.

A multipeak-capable laser profile scanner offers a configurable maximum peak count per pixel column, typically selectable from 1 (standard single-peak mode), 2, 3, or 4 peaks. Increasing the maximum peak count increases the signal processing time per profile line, reducing the maximum measurement rate:

Multipeak evaluation is implemented either as on-sensor processing — running on the sensor’s FPGA or DSP before data transmission, reducing interface bandwidth requirements — or as host-side post-processing after raw image transmission; on-sensor multipeak processing transmits only the extracted peak data rather than full raw images.

Multipeak detection is one of 3 advanced signal evaluation features that operate within the same Lasertriangulation node. HDR (High Dynamic Range) extends the measurable reflectivity range by combining multiple exposures to simultaneously evaluate very dark and very bright surface areas — a capability complementary to multipeak detection on mixed-reflectivity surfaces. Multiple Slope adapts the sensor’s exposure behavior at steep surface angles and strong intensity gradients, improving signal quality at geometry edges where interreflection ghost peaks frequently arise. Multipart evaluation enables the sensor to assign pixel columns to multiple independent measurement regions within a single scan pass. Each of these 3 features has its own dedicated article within the Lasertriangulation node.

Multipeak output multiplies the profile data volume proportionally to the configured peak count. A sensor operating at 4,000 profiles per second with 1,280 pixel columns in single-peak mode transmits 5.12 million z-values per second. In 2-peak mode the same sensor transmits 10.24 million z-values per second; in 4-peak mode it transmits 20.48 million z-values per second. Industrial interface standards used with laser profile sensors — including GigE Vision(bandwidth: up to 125 MB/s), Camera Link Full (bandwidth: up to 680 MB/s), and CoaXPress(bandwidth: up to 12.5 GB/s per lane) — determine whether the increased data volume of multipeak mode can be sustained at the required measurement rate. Downstream processing hardware and software must be configured to handle the expanded data structure, including per-peak amplitude and validity fields, before multipeak mode can be deployed in a production inspection system.

The following topics are adjacent to multipeak detection and are covered in separate dedicated articles. The Lasertriangulation parent article covers the geometric measurement principle, Scheimpflug condition, and calibration methodology — these fundamentals are not repeated here.HDR covers dynamic range extension for simultaneous measurement of high-contrast surface areas. Multiple Slope covers multi-exposure signal adaptation at steep surface angles. Multipartcovers multi-region evaluation within a single profile scan. The CCD/CMOS Sensor component article covers image sensor architecture, pixel structure, and readout electronics. The Reflexion/Absorption measurement principle article covers the optical physics of surface reflection and absorption. Layer-thickness derivation from the peak centroid separation of a 2-peak measurement on a transparent material is covered in the dedicated metrology article on layer-thickness measurement. Laser safety classification and laser class markings are not within the scope of this article.