Decorative wall surface flatness detection system based on computer vision

Abstract

The invention relates to the field of constructional engineering quality detection and computer vision, in particular to a decoration wall flatness detection system based on computer vision. The device is composed of a processor, a double-conical-surface projection unit, a wave band separation unit, an event image acquisition unit, a synchronous control unit, a conical surface parameter solving unit, a scale self-calibration unit, a depth reconstruction unit, a flatness evaluation unit and a result generation unit. And a conical surface parameter is obtained through analysis, and a pixel-length scale factor is solved in a closed mode. The method comprises the following steps: establishing a pixel-physical length absolute scale by self-calibration based on a closed scale derived from a known distance between vertexes of two conical surfaces and an opening angle through a cooperative scheme of double-waveband double-conical-surface structured light projection synchronization time coding gating and event image acquisition; and the depth of the wall surface is stably reconstructed in single scanning, and flatness and a traceable report are output, so that the influence of external calibration and crosstalk is eliminated under strong ambient light and complex reflection.

Description

Technical Field

[0001] This invention relates to the fields of building engineering quality inspection and computer vision, and in particular to a computer vision-based system for detecting the flatness of decorated walls. Background Technology

[0002] In interior decoration projects, wall flatness acceptance is often based on millimeter-level tolerance. There are three existing technologies: first, manual point measurement (using rulers and feeler gauges), which is inefficient, subjective, and difficult to quantify; second, camera-laser / structured light 3D measurement, which requires prior calibration, and on-site temperature changes and displacement can easily cause calibration failure; and third, binocular / SLAM photogrammetry, which is easy to deploy but lacks features on walls with weak textures, and requires additional measuring tools for absolute dimensions.

[0003] Existing structured light solutions rely on calibration boards or pre-calibration to establish mapping. Strong ambient light (outdoor light, floodlights, power line flicker) or complex reflective surfaces (coexistence of high / low reflectivity, highlights, dark seams, roughness) can easily cause overexposure / underexposure and decreased stripe contrast in frame cameras, leading to decoding errors and depth holes. Single-band projection in mixed reflective scenes is prone to superimposed with the background spectrum, causing crosstalk; if superimposed with micro-vibrations or occlusion, data contamination will reduce the reliability of the assessment.

[0004] Project acceptance requires full-area absolute dimensional results and a traceable report. Traditional processes require carrying a calibration board (reducing mobility) or only obtaining a relative height map (requiring manual conversion), making it difficult to consistently produce "millimeter-level, absolute dimensional, and verifiable" indicators in a single scan. Large-area wall surfaces require rapid scanning and calculation, and existing solutions that require repeated focusing and calibration cannot meet the needs of on-site efficiency and stability.

[0005] In summary, the technical problem actually solved by this invention is: how to stably establish the absolute scale of "pixel-physical length" in a single scanning cycle in a decorative wall scene with strong ambient light and complex reflections, without the need for external calibration plates and on-site measuring tools, and obtain millimeter-level flatness results and traceable reports for engineering acceptance. Summary of the Invention

[0006] To overcome the aforementioned technical deficiencies, the present invention aims to provide a computer vision-based wall surface flatness detection system. This invention employs a collaborative scheme of "dual-band dual-cone structured light projection + synchronous time-coding gating + event image acquisition," and establishes an absolute scale of pixel-physical length based on the known distance s between the vertices of the two cones and the opening angle through closed-scale self-calibration. This allows for robust reconstruction of wall surface depth within a single scan and output of flatness and traceable reports, thereby eliminating the influence of external calibration and crosstalk under strong ambient light and complex reflections.

[0007] This invention discloses a computer vision-based system for detecting the flatness of a decorated wall surface, comprising: an event image acquisition unit, a dual-cone projection unit, a band separation unit, a synchronization control unit, a cone parameter solving unit, a scale self-calibration unit, a depth reconstruction unit, a flatness evaluation unit, a result generation unit, a storage unit, and a processor; wherein the processor and storage unit are electrically connected, and are signal-connected to the event image acquisition unit, dual-cone projection unit, band separation unit, synchronization control unit, cone parameter solving unit, scale self-calibration unit, depth reconstruction unit, flatness evaluation unit, and result generation unit, respectively; the dual-cone projection unit is used to simultaneously project a first non-coplanar cone surface and a second non-coplanar cone surface onto the wall surface being measured within a single scanning cycle, wherein the opening angle of the first cone surface is [missing information]. The opening angle of the second cone is The vertices of the two cones are a fixed, known distance apart along the projection optical axis. The opening angle , A positive quantity expressed in radians or degrees. The quantity is a positive value expressed in units of length; the band separation unit is used to position the first and second conical surfaces in non-overlapping first and second bands, respectively, and to provide corresponding bandpass channels on the imaging side; the event image acquisition unit is used to record the event stream and its timestamp sequence triggered by the first and second conical surfaces within a single scan cycle. The synchronization control unit is used to perform hardware-triggered synchronization between the event image acquisition unit and the dual-cone projection unit, and to align the time index of the projection command sequence with the event timestamp; the cone parameter solving unit is used to solve the analytical parameters of the first and second cones in the camera coordinate system, and the analytical parameters include at least their respective opening angle, vertex position, and axial direction; the scale self-calibration unit is used to determine the scale based solely on known distances. Opening angle , And the geometric quantities of the intersection lines of the two conical surfaces and the wall surface on the image plane, to calculate the pixel-length scaling factor. And establish the absolute scale of the wall surface, pixel-to-length ratio factor. Used to convert pixel distances in an image into physical length; the depth reconstruction unit is used for each pixel Calculate the two sets of heights given by the first and second cone surfaces. and The system outputs the fusion height and effective pixel mask; the flatness evaluation unit is used to perform planar fitting on the fusion height in the indoor coordinate system and output the flatness index of the evaluation grid; the result generation unit is used to output the overlay layer registered with the wall image and the data file containing the position index and statistics.

[0008] Compared with existing technologies, the above technical solution has the following advantages:

[0009] 1. Absolute dimensional accuracy can be established without external calibration. Existing technologies generally rely on external calibration plates / gauges or pre-calibrated internal and external parameters. On-site temperature drift, focus changes, and adjustments can lead to inaccuracies. These methods are inefficient in terms of handling and placing calibration plates, require repeated calibration during retesting, and are susceptible to environmental disturbances. In this invention, the double-cone projection geometry and known spacing are utilized... Opening angle , In the self-calibration unit, a closed-form scaling factor is given through the intersection of two coaxial loops. ( and (The number of pixels measured along the same line of the image plane) allows for obtaining absolute scale within a single scan cycle without relying on any external calibration objects, resulting in faster deployment and more stable retesting.

[0010] 2. Data purity under strong ambient light and complex reflections. In existing technologies, single-band stripes / laser lines are easily obscured by construction site floodlights, sunlight, and fluorescent light flicker; the mixture of white walls / high reflectivity / dark seams leads to both overexposure and underexposure, resulting in high crosstalk, numerous misinterpretations, and many holes. In this invention, dual-band beam splitting + polarization multiplexing + equivalent neutral density (ND) + time gating are used; the crosstalk rate is defined by the band separation unit. , It also features closed-loop suppression; the event image acquisition unit receives events within the gated window, significantly reducing background triggering and cross-channel contamination, and can still stably acquire effective data in strong light / flicker scenarios.

[0011] 3. High dynamic range and motion robustness. In existing technologies, frame-based cameras are prone to overexposure / underexposure, motion blur caused by movement / shaking, loss of detail in dark slits / highlights, edge blurring, and measurement instability. This invention addresses these issues by employing event-based image acquisition (triggered by logarithmic brightness changes + pixel refractory period). With microsecond-level timestamps and gating from the synchronous control unit, it achieves inherently high dynamic range and zero / near-zero motion blur, making it less sensitive to vibration and personnel movement. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating a computer vision-based wall surface flatness detection system according to the present invention.

[0013] Figure 2 A schematic diagram of the pixel triggering model for the event image acquisition unit;

[0014] Figure 3 Gating function for synchronization control unit , Schematic diagram of phase misalignment;

[0015] Figure 4A schematic diagram of the dense height map contour lines generated for the depth reconstruction unit;

[0016] Figure 5 This is a schematic diagram of the grid and residuals for the flatness evaluation unit. Detailed Implementation

[0017] The advantages of the present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments.

[0018] See Figure 1 As shown in the figure, this embodiment provides a computer vision-based wall surface flatness detection system, including: an event image acquisition unit, a dual-cone projection unit, a band separation unit, a synchronization control unit, a cone parameter solving unit, a scale self-calibration unit, a depth reconstruction unit, a flatness evaluation unit, a result generation unit, a storage unit, and a processor.

[0019] The processor will be described in detail in this embodiment. The processor includes: a time base management submodule, an acquisition interface submodule, a control and synchronization submodule, a cone parameter solving submodule, a scale self-calibration submodule, a fusion reconstruction submodule, an evaluation and coordinate alignment submodule, an anomaly detection and self-recovery submodule, and a storage and output submodule. The processor is connected to the event image acquisition unit via a first high-speed data bus to receive event streams and timestamps; it is connected to the dual-cone projection unit via a second control bus to send drive frequencies and phases and receive status readbacks; it is connected to the synchronization control unit via a third synchronization signal line to send and receive hardware trigger, time base, and gating window opening commands; it is connected to the band separation unit via a fourth data channel to read dual-band channel counts, bandpass status, and signal-to-noise ratio estimates; it is connected to the indoor coordinate establishment unit via a fifth data channel to exchange the wall-to-ground intersection direction and gravity direction estimation vectors; and it is connected to the result generation unit via a sixth data channel to output overlay layers and statistical data files. The processor is also software-coupled and invoked within the cone parameter solving unit, scale self-calibration unit, depth reconstruction unit, and flatness evaluation unit.

[0020] After the system powers on, the time base management submodule establishes a globally monotonically increasing clock and distributes time signals to various peripherals; the acquisition interface submodule performs unified time base recalibration on the event timestamps output by the event image acquisition unit; and the control and synchronization submodule performs joint configuration of the biconical projection unit and the synchronization control unit according to predetermined constraints. The mathematical quantities involved are defined as follows:

[0021] The first driving frequency of the dual-cone projection unit is denoted as: (Hz), the second driving frequency of the double-cone projection unit is denoted as (Hz); the first time-encoded phase offset of the synchronization control unit is denoted as (rad), the second time-coded phase offset of the synchronization control unit is denoted as (rad); the first time encoding period of the synchronization control unit is denoted as (s), the second time encoding period of the synchronization control unit is denoted as (s); The time resolution period of the event image acquisition unit is denoted as (s); The gate half-width setting value of the synchronization control unit is denoted as (s). The processor is used to write the following constraints: and used to restrict and , The relationship satisfies the coprime period and gated window width constraints to ensure non-aliasing untangling within the same scan period.

[0022] The processor is used to execute time consistency and deskipation processes; the cross-correlation function of the dual-channel event density sequence output by the event image acquisition unit is denoted as... (Independent variable s), the time bias to be estimated is denoted as (s), the processor is used to solve ,in This represents the variable that minimizes the objective function; and is used to... Write back to the synchronization control unit to correct trigger alignment. Experimental results: After ten repeated measurements at room temperature, using the processor's time deskewing process, the median absolute deviation of the dual-channel event arrival time decreased from 38 microseconds to 9 microseconds, and the longitudinal jitter of the depth reconstruction unit output decreased by approximately 52% (1 ).

[0023] The processor is used to perform joint solution of conical surface parameters and observability monitoring; the opening angle of the first conical surface of the biconical projection element is denoted as... (rad or angle), the opening angle of the second cone of the double-cone projection unit is denoted as (rad or angle), the known distance between the vertices of the two cones of the biconical projection unit along the projection optical axis is denoted as... (m); The Jacobian matrix mapping from the cone parameters to the image plane observation is denoted as... The Jacobian condition number threshold is denoted as The processor is used to calculate the condition number. ,when This is used to adjust the phase offset of the synchronization control unit. , With the driving frequency of the double-cone projection unit , To reduce the severity of the illness. Data support: When At that time, the pathological solution trigger rate was less than 3%, and the standard deviation of the three-dimensional height decreased from 0.92 mm to 0.54 mm (n=15).

[0024] The processor is used to perform scale self-calibration and online robust updates; where the pixel-to-length scaling factor is denoted as... (m / pixel), the previous period's scaling factor estimate is denoted as (m / pixel), the resolution scaling factor for this period is denoted as (m / pixel), global update coefficient denoted as (Dimensionless), mass fraction is denoted as (Dimensionless, derived from the signal-to-noise ratio and event density assessment of the band separation unit), the online weight is denoted as... The processor is used for... , , The geometric quantities of the intersection line of the two cone surfaces and the wall surface on the image plane are obtained analytically. And used for pressing Online smooth updates are performed to establish an absolute scale; comparative data: under temperature variations of 20–40℃, the pixel-to-length scaling factor after online updates is used. The relative error is stabilized within 0.3%, which reduces longitudinal scale drift by about 65% compared to the fixed-scale method.

[0025] The processor is used to perform pixel-level consistency detection and fusion reconstruction; where the depth reconstruction unit is at the pixel level. The height obtained from the first cone surface is denoted as (m), the height obtained from the second cone surface is denoted as (m), the adaptive consistency threshold is denoted as (m), the basic threshold setting value is denoted as (m), the proportional coefficient setting value is denoted as (Dimensionless), the fusion height is denoted as (m), pixel weight denoted as and (dimensionless and satisfying) The processor is used for calculation. and in Time output This is used to mark a pixel as invalid when it does not meet a threshold or when the quality metric exceeds the limit. Results: On both gray-white putty walls and lightweight brick walls, the proportion of outliers decreased from 2.8% to 0.6%, and the maximum residual in a 1m×1m area decreased from 1.8mm to 0.9mm.

[0026] The processor is used to perform evaluation mesh adaptation and indoor coordinate alignment; the side length of the evaluation mesh is denoted as L (m), and the mesh scale factor setting value is denoted as... (Dimensionless), the interval-limited function is denoted as The processor is used to determine the appropriate parameters based on the given information. The grid scale is set and used to complete coordinate alignment under the constraints of the wall-ground intersection direction vector and gravity direction vector provided by the indoor coordinate establishment unit. It is then used to perform weighted least squares plane fitting in each grid and output a weighted flatness index. Experimental retest: The test of the corridor wall prototype with 24 grids showed that the number of grid condition out-of-bounds events decreased by 72%, and the evaluation repeatability improved from 0.62mm to 0.38mm.

[0027] The interaction between the processor and the storage unit is as follows: the processor loads parameter configurations (including but not limited to the aforementioned) from the storage unit. (etc.), and is used to archive fusion height, weighted flatness index, duty cycle vector, pixel-length scaling factor online trajectory, Jacobian condition number time series and key intermediate quantities to storage units for traceability verification and quality control.

[0028] It should be noted that the processor is also used to calculate the time bias based on cross-correlation and write back the synchronization control; the processor is used to drive adaptive parameter tuning of frequency and phase using condition number thresholds; and the processor is used to apply online weights. Smoothly update pixel-length scaling factor The processor is used to adaptively threshold... Guide pixel-level blending; processor used for pixel-level blending The evaluation grid is adaptively set; the processor is used to perform temperature-frequency joint compensation, low duty cycle anomaly detection, geometric observability assurance, and anomaly self-recovery and re-sampling.

[0029] This embodiment will describe the dual-cone projection unit in detail. The dual-cone projection unit includes: a light source submodule, a band beam combining submodule, an acousto-optic deflector module, a coaxial image relay lens group, an axial cone lens assembly, a ring energy shaping submodule, an exit port and safety shutter submodule, and a temperature control and status monitoring submodule. The dual-cone projection unit is connected to the processor via a control and status bus to receive drive frequency, phase, and safety control commands and to report temperature, power, and alignment status. The two outgoing beams of the dual-cone projection unit correspond spectrally to the imaging-side bandpass channel of the band separation unit, which is used to achieve consistency in channel differentiation and event indexing.

[0030] The light source submodule is used to output collimated beams in two independent wavelength bands; the center wavelength of the first wavelength band is denoted as . (nm), the center wavelength of the second band is denoted as (nm), the difference in center wavelength between the two bands is denoted as (nm), Used to meet the requirements of imaging side bandpass separation and preferably meet The lower limit, of which This is the design lower limit setting; the emitted power of the light source in the two bands is denoted as... and (mW) is used to provide a sufficient signal-to-noise ratio while meeting the safety requirements.

[0031] The band beam combiner module is used to combine two band beams onto the same optical axis; a dichroic beam combiner is used to achieve beam combining, and the effective aperture of the combined beam is limited by an aperture. The angular deviation of the beam combining error is denoted as... (Unit: milliradians), the lateral deviation of the beam combining error is denoted as... (mm) and Used as an evaluation metric for assembly and adjustment, and periodically collected and reported by the temperature control and condition monitoring submodule.

[0032] The acousto-optic deflector module is used to simultaneously modulate and deflect two beams of light in the same period; the first driving frequency is denoted as... The second driving frequency is denoted as (Hz), the frequency sensitivity of the acousto-optic deflection angle is denoted as (rad / Hertz), the deflection relationship based on the linear small-angle approximation is as follows: ,in Indicates the instantaneous deflection angle (rad). Indicates the current driving frequency (Hz). The center frequency (Hz) of the acousto-optic device is represented; the time-encoding phase offsets of the two channels are denoted as... , (rad), together with the synchronization control unit, is used to form the simultaneous projection condition of "same period, independent encoding", preferably satisfying , ,in , These represent the time coding periods (s) of the two channels, respectively. This represents the time resolution period (s) of the event acquisition link. The coaxial image relay mirror group is used for pupil plane image relay and axial conjugate control of the combined beam, thereby applying a dual-parameter constraint of "vertex position-aperture angle" to the two projection geometries; the mechanical spacing of the coaxial image relay mirror group is denoted as... (mm), axial magnification is denoted as (Dimensionless), the design distance between the vertices of the two conical surfaces along the projection optical axis is denoted as... (mm), satisfying ,in The residual spacing error (mm) is indicated; the tilt tolerance of the coaxial relay lens group is denoted as... (rad) is used to limit the coaxiality of the two beams. With the above constraint, the biconical projection unit forms two independent and coaxial cone vertices in space.

[0033] The axial-cone lens assembly is used to convert the annular energy in the pupil plane into the exit geometry of the "first cone" and the "second cone"; where the opening angle of the first cone is denoted as... (rad or angle), the opening angle of the second cone is denoted as (rad or angle), the base angle parameters of the axial cone lens are denoted as follows: and (rad or angle), the refractive index of the lens material is denoted as (Dimensionless), under the small-angle approximation, the design relationship between the opening angle and the base angle can be written as: , ,in This represents the dimensionless angular magnification factor introduced by the coaxial image relay lens group. For ease of system joint calibration, the biconical projection unit preferably satisfies the upper limit constraint of the aperture angle difference. ,in This indicates the upper limit (rad or angle) of the opening angle difference.

[0034] The annular energy shaping submodule is used to homogenize the annular cross-section and suppress side lobes at the exit port; wherein the intensity uniformity of the annular cross-section is indicated as follows: (Dimensionless, defined as the ratio of the difference between the maximum and minimum values ​​to the average value), the target values ​​for the annular thickness are denoted as follows: and (mm) , , Used as a quality metric and acquired by the processor cycle; when Exceeding the set threshold or , When deviating from the target range, the processor is used to adjust. , and , Subtle disturbances to reduce the fixed pattern.

[0035] The exit port and safety shutter submodule are used to deliver the two-conical beam into the measured area and quickly cut it off in case of an anomaly; the closing time of the safety shutter is denoted as... (ms), the over-temperature threshold of the temperature control and status monitoring submodule is denoted as (°C), the upper limit of the output power is denoted as (mW); when detected or or Or bundle combination error When the time limit is exceeded, the safe shutter speed is... The system is closed internally to ensure safety and reports the status to the processor and result generation unit.

[0036] To establish the geometric relationship between the "cone surface and spatial position", the axial distance of the reference coordinates along the projection optical axis is defined as... (m, the origin is located at the corresponding vertex of the cone), under free propagation conditions, the two cones and the plane The radii of the intersection lines respectively satisfy , ,in , These represent the distances between the first and second conical surfaces, respectively. The annular radius (m) at that location. The above relationship matches the analytical equations of the conical parameter solution element and provides observable geometric quantities for the scale self-calibration element.

[0037] Regarding the assembly and self-inspection process, the dual-cone projection unit is used to perform a three-step self-inspection: First, the temperature control and status monitoring submodule heats the temperature at a constant slope and records the temperature before steady-state establishment. The drift curve is used to generate the temperature drift compensation coefficient; secondly, the coaxial relay mirror group is used to output the conical pattern for self-labeling, and the event image acquisition unit works with the processor to... , and Make an estimate, if The prompt indicates fine-tuning; thirdly, the ring energy shaping submodule is used to output a standard ring, and the band separation unit measures on the two bandpass channels. , , If it is outside the tolerance range, the processor will use it to send frequency and phase perturbations (amplitude denoted as...). (rad) to reduce fixed speckle.

[0038] In terms of connectivity and interaction, the dual-cone projection unit receives data through: the first control register group. , Target value and safety limit Over-temperature threshold The second control register group receives... , With time encoding period , Status register group reporting The fast channel is used to receive the safety shutter closing command and... It is executed internally; all register mappings are managed uniformly by the processor and kept in sync with the synchronization control unit according to a unified time base.

[0039] Regarding the coordination with other units in the system, the dual-cone projection unit is used for: collaborating with the synchronization control unit to generate simultaneous encoding that satisfies the constraints of coprime periods and phase reversal; maintaining strict spectral correspondence with the band separation unit to achieve consistency in the "projection-imaging" channel; maintaining temporal non-aliasing and untangling with the event image acquisition unit; and sharing resources with the cone parameter solving unit. , The initial value is obtained and improved through iteration; it is shared with the scale self-calibration unit. Online observation or estimation; improve the pixel-level effective mask ratio with the flatness evaluation unit through geometric consistency; share quality metrics and safety event logs with the result generation unit to form a traceable record.

[0040] Through the above structure and process, the dual-cone projection unit stably forms a dual-cone projection geometry under the conditions of "simultaneous and simultaneous projection with the same period, separation of the two bands, controlled vertex spacing and settable opening angle". The geometric stability is guaranteed by quantifiable assembly tolerances and in-situ monitoring closed loop, providing high-quality priors and observations for the system's cone parameter solving unit, scale self-calibration unit and depth reconstruction unit, thereby obtaining reproducible flatness evaluation results in large flat scene such as wall.

[0041] See Figure 2 As shown, the event image acquisition unit will be described in detail in this embodiment. The band separation unit includes: an incident window submodule, a dichroic beam splitting submodule, an image plane bandpass subchannel, an image plane microfilter array submodule, a polarization selection submodule, an angular incident compensation submodule, a stray light suppression and equivalent neutral density adjustment submodule, a channel equalization and quality assessment submodule, and a status monitoring and temperature control submodule. The band separation unit is connected to the processor through a first data channel and is used to report spectral and channel quality measurements and receive adjustment commands. The band separation unit and the event image acquisition unit together form an image-side acquisition link through the imaging optical path, which is used to achieve pixel-level band differentiation between the "first cone" and the "second cone" on the imaging side. The band separation unit strictly corresponds to the two bands output by the biconical projection unit in the spectrum to ensure the consistency of the projection-imaging channel.

[0042] The incident window submodule is used to provide anti-reflection and environmental sealing; the reflectivity of the first surface is denoted as... (Dimensionless), the reflectivity of the second surface is denoted as (Dimensionless), the upper limit of the targets for both is denoted as... (dimensionless); when or At this time, the status monitoring and temperature control submodule is used to report deviations and trigger self-tests.

[0043] The dichroic beam splitter module is used to split the incident light into two image plane bandpass channels according to wavelength ranges; the center wavelength of the first split bandpass is denoted as . (nm), the center wavelength of the second flow band split is denoted as (nm), the difference between the two center wavelengths is denoted as (nm), the full width at half maximum (FWHM) of the two passbands are denoted as (nm) and (nm) respectively. or (nm); the slope of the beam splitting edge is denoted as Used to limit the separation sharpness; the angle of incidence of the dichroic photonic module is (Units are degrees or radians), under the conditions of interferometric devices, its center wavelength angular drift satisfies ,in Represents the equivalent refractive index (dimensionless). The angular incidence compensation submodule is used to... Adjust the installation posture in time to maintain , Stablize.

[0044] The image plane bandpass sub-channel is used to construct two bandpass responses on the imaging side; where the first bandpass transmittance function is denoted as... (Dimensionless), the second-band transmittance function is denoted as (Dimensionless), the spectral response function of the event image acquisition unit is denoted as... (Dimensionless); The spectral power densities of the biconical projection element in the two bands are denoted as follows: and (Unit: power / nanometer). To measure spectral crosstalk, the spectral crosstalk rate of the two channels is defined. , ,in This represents the definite integral with respect to wavelength. The channel equalization and quality assessment submodule is used for real-time calculation. , and the upper limit (Dimensionless) comparison, when At this time, the processor is used to issue adjustment instructions, including fine-tuning the bandpass center, adjusting the polarization state, or changing the gating strategy.

[0045] The image-plane microfilter array submodule is used to implement pixel-level band labeling; the tiling units of the microfilter array are repeating in a checkerboard pattern, and the fill factor of the first sub-pixel is denoted as . (Dimensionless), the fill factor of the second sub-pixel is denoted as (Dimensionless, defined at the corresponding sub-pixel); the event image acquisition unit reconstructs channel labels according to the pixel neighborhood, and the processor divides the events into a first event set and a second event set accordingly. To measure index accuracy, the band index misclassification rate is defined. ,in This represents the number of events that were incorrectly assigned to another frequency band (dimensionless count). The total number of events (dimensionless count) is represented by the channel equalization and quality assessment submodule, which is used for output. and with threshold Compare.

[0046] The polarization selection submodule is used to introduce mutually orthogonal polarization states in two bands to reduce crosstalk caused by multiple reflections; the polarization direction angle of the first band is denoted as... (Units are degrees or radians), the polarization direction angle of the second band is denoted as... (Units are degrees or radians), preferably satisfying The extinction ratio of the polarizer group is denoted as... (Dimensionless), used to measure polarization selectivity. The polarization selection submodule is linked to the dichroic spectrophotometer to achieve polarization selectivity without changing the polarization selectivity. , In the case of reducing , .

[0047] The stray light suppression and equivalent neutral density adjustment submodule is used to reduce background triggering in high-illuminance or strong ambient light scenarios; where the background spectral power density is denoted as... (Unit: power / nanometer), equivalent neutral density value is denoted as ND (dimensionless, expressed as log transmittance), background suppression ratio is denoted as... (Dimensionless, defined as the ratio of the background event rate after suppression is enabled to the background event rate before suppression is enabled); when When the target lower limit is reached, the processor is used to increase the equivalent neutral density of the two bands or narrow the full width at half maximum (FWHM) of the bandpass. , .

[0048] The angle incidence compensation submodule is used to compensate for the angular drift of interferometric bandpass under conditions of large aperture and large field of view; where the image height is denoted as... (mm), the angle between the camera's principal ray and the normal is denoted as . (Units are degrees or radians), and the resulting center wavelength offset is denoted as follows: and (nm); The angle incidence compensation submodule adopts a combination of a micro-tiltable mounting ring and a field curvature compensation thin wedge, so that... Less than the target upper limit .

[0049] The channel equalization and quality assessment submodule is used to perform online equalization of energy and signal-to-noise ratio in two bands; where the channel gain is denoted as... and (Dimensionless), event density is denoted as and (Units are events / pixels / periods), the target event density is denoted as... (Units as before), the processor is used to decode To obtain , The online adjustment amount is used to adjust the gated half-width setting value in the synchronization control unit. Suppressing oversaturation triggering through synergy.

[0050] In terms of connectivity and interaction, the band separation unit receives the target center wave through: the first control register group. , Half height and full width , Equivalent neutral density ND and polarization direction , The second control register group sets the tilt angle and thin wedge for incident compensation of the receiving angle; the status register group reports... (in (Indicates the current device temperature, °C); all registers are managed uniformly by the processor and share the quality log with the result generation unit.

[0051] Regarding its coordination with other system units, the band separation unit is used to: maintain correspondence with the biconical projection unit in terms of center wavelength and full width at half maximum (FWHM) parameters, so that... , and , Matching; with the synchronization control unit at the gate half-width setting. With time encoding period , A joint "spectral-temporal" unwrapping is formed; pixel-level tag transfer is achieved with the event image acquisition unit through the image plane microfilter array submodule; and it shares with the processor. , , , , , Equal mass quantities drive online equilibration; the reliability of geometric quantity observations is improved by reducing cross-contamination with the cone parameter solving unit and the scale self-calibration unit; and channel quality logs are shared with the result generation unit to form traceable records.

[0052] Through the above structure and process, the band separation unit, under the combination of "dichroic spectral dispersion + image plane bandpass + microfilter pixel label + polarization selection + angular incidence compensation + online equalization", stably achieves pixel-level band differentiation of the two conical projection geometry, and uses quantized spectral crosstalk rate, index misjudgment rate and background suppression ratio as quality indicators.

[0053] This embodiment will describe the event image acquisition unit in detail. The event image acquisition unit includes: an imaging window submodule, a pixel array and photoelectric front-end module, a polarity separation and threshold comparison submodule, a timestamp engine submodule, an address-event representation bus submodule, a data packing and buffering submodule, a gating and region selection submodule, a synchronization interface submodule, and a calibration and self-monitoring submodule. The event image acquisition unit is connected to the processor through a first high-speed data bus for transmitting event streams and status registers; it is connected to the synchronization control unit through a first synchronization trigger line for receiving global trigger, gating windowing, and time base alignment signals; optically, it is coaxially arranged with the band separation unit to achieve pixel-level band differentiation between the two image plane bandpass channels.

[0054] The imaging window submodule is used to provide anti-reflection and environmental sealing; the reflectivity of the first surface is denoted as... (Dimensionless), the reflectivity of the second surface is denoted as (Dimensionless), the upper limit of both targets is denoted as... (dimensionless), when or At that time, the calibration and self-monitoring submodule is used to report deviations and trigger self-checks.

[0055] The pixel array and optoelectronic front-end module are used to convert incident light into electrical signals; the number of pixels in the horizontal direction of the array resolution is denoted as... (Dimensionless counting), the number of pixels in the vertical direction is denoted as (Dimensionless counting), pixel pitch is denoted as (Unit: micrometers), pixel full-well charge is denoted as (Unit: coulomb), the equivalent input reference current is denoted as... (Unit: Ampere). The event image acquisition unit preferably has high dynamic range imaging capability, typically above 120 dB, to adapt to the on-site differences between "strong reflection from white walls and weak reflection from dark seams".

[0056] The polarity separation and threshold comparison submodule is used to asynchronously generate "on" and "off" events at the pixel level; to describe the triggering rules, pixels are defined. At the moment light intensity is (Unit: optical power), the logarithmic intensity is defined as... (Dimensionless), defines a pixel The last trigger time was (s), define the "start event threshold" as (Dimensionless), define the "close event threshold" as: (Dimensionless). The polarity separation and threshold comparison submodule is used to determine the polarity separation and threshold comparison conditions. Output start event when the condition is met The system outputs a shutdown event and hands over the polarity, pixel coordinates, and timestamp to the timestamp engine submodule for processing. To suppress "jitter triggering," the polarity separation and threshold comparison submodule applies a pixel refractory period after each trigger, denoted as . (Unit: microseconds).

[0057] The timestamp engine submodule is used to assign a timestamp to each event using a uniform time base; where the time resolution period is denoted as... (s), the time quantization step is denoted as (s, The standard deviation of timestamp jitter is denoted as (s). The synchronization interface submodule is used to complete time base alignment under the timing of the synchronization control unit. The timestamp engine submodule performs correction after receiving the alignment amount to ensure the consistency of cross-channel event time on the processor side.

[0058] The gating and region selection submodule is used to perform time window gating and spatial region selection; where the gating window width is denoted as... (s), the time coding period is denoted as and (s), the gating and region selection submodule is used to select regions only when the width is The corresponding channel events are accepted within the valid window, and events are suppressed outside the window; the set of interest regions selected in the spatial region is denoted as... (A set indexed by pixel coordinates), the gating and region selection submodules are used to select only regions... Intra-pixel reporting events, It can be updated online by the processor based on quality metrics.

[0059] The address-event representation bus submodule is used to output events asynchronously without a shared clock; where the maximum sustainable event rate is denoted as... (Unit: events / second), instantaneous packet loss rate is denoted as... (Dimensionless, defined as the ratio of the number of events that failed to enter the bus to the total number of events), the data packaging and buffering submodule is used to package events into fixed time slices, the width of which is denoted as... (ms), and temporarily buffered when the bus is congested, with the buffer capacity denoted as . (Unit: events). The data packaging and buffering submodule is also used to schedule events according to channel priority and polarity polling strategies, with channel priority weights denoted as... , (dimensionless, satisfies) ).

[0060] The calibration and self-monitoring submodule is used for online maintenance of pixel-level thresholds, pixel refractory periods, and thermal noise suppression, and periodically generates quality metrics; the count of on-event events and the count of off-event events are denoted as follows: (Dimensionless counting), polarity balance is denoted as (Dimensionless), the density of the target event is denoted as (Units are events / pixels / periods), the current event densities of the two channels are denoted as follows: and (Units as before), the upper limit of the noise event rate is denoted as... (Units are events / pixels / second). The calibration and self-monitoring submodule is used to adaptively update pixel-level thresholds using the following formula: , ,in This represents the threshold update coefficient (dimensionless). , These represent the event density for turning on and off (units are events / pixel / period); when the pixel noise event rate exceeds... At that time, the calibration and self-monitoring submodule is used to increase the pixel refractory period. The pixel is then reported as a "hot pixel," and the processor subsequently decides whether to include it in the region of interest set. Remove from the list.

[0061] To illustrate the time-spectral consistency of channel differentiation, the first channel gating function is defined as follows: (Dimensionless), the second channel gating function is (Dimensionless), they are respectively controlled by the synchronization control unit according to the time coding period. , With window width The generation; gating and region selection submodule is used to receive events according to the following rules: when the pixel bandpass label belongs to the first bandpass and Receive event when the pixel bandpass label belongs to the second bandpass and Receive events, otherwise suppress them; this rule is consistent with the pixel-level bandpass label of the band separation unit to reduce cross-channel aliasing.

[0062] In terms of connectivity and interaction, the event image acquisition unit receives data through: the acquisition control register group. , , , With interest area set Time-gated register group reception , , Global synchronization pulse; status register group reporting , , , , , Along with hot pixel counting; all registers are managed uniformly by the processor and share a quality log with the result generation unit to form a traceable record.

[0063] Regarding its coordination with other system units, the event image acquisition unit is used to: complete the consistency transmission of bandpass tags at the pixel level with the band separation unit; and coordinate with the synchronization control unit during the time coding cycle. , With door control window width A time-spectral joint unwrapping is formed; shared with the processor , , , , Equal mass quantities drive time deskipation, threshold adaptation, and region of interest adjustment; with the cone parameter solving unit, the condition number stability of parameter estimation is improved through time alignment and noise suppression; with the scale self-calibration unit, the reliability of geometric observations is improved by reducing cross-channel aliasing; and with the results generation unit, a quality log is shared to form a traceable record.

[0064] Through the above structure and process, the event image acquisition unit, with the combination of "pixel-level logarithmic change triggering + unified time base for timestamps + gating and region selection + high-speed asynchronous output + online thresholding and thermal noise suppression", can stably provide two-channel, low crosstalk, and low packet loss event streams in strong dynamic range and complex lighting environments.

[0065] See Figure 3 As shown in this embodiment, the synchronization control unit will be described in detail. The synchronization control unit includes: a main time base and timing submodule, a triggering and phase scheduling submodule, a time encoding and gating generation submodule, a jitter suppression and phase servo submodule, a frequency drift compensation submodule, a health monitoring and watchdog submodule, and a register and event log submodule. The synchronization control unit is connected to the event image acquisition unit through a first synchronization signal line for issuing global triggers, timing pulses, and gating windows; it is connected to the biconical projection unit through a second control bus for distributing time-coded phase and frequency sequences; it is connected to the processor through a third data channel for receiving correction quantities, quality measurements, and parameter tuning commands and transmitting the status back; and it maintains a consistent time-spectrum correspondence with the band separation unit in terms of gating strategy, thereby achieving "simultaneous, independent encoding, and non-aliasing" two-channel operation within the same scanning cycle.

[0066] The main time base and timing submodule is used to generate the system reference clock; the system reference clock frequency is denoted as . (Hz), the standard deviation of short-term clock jitter is denoted as (s), the long-term drift rate is denoted as (s / second); The main time base and timing submodule are used to provide unified timing for the event image acquisition unit and the dual-cone projection unit through timing pulses, ensuring that all peripherals work under the same monotonically increasing time base.

[0067] The triggering and phase scheduling submodule is used to organize all time events within the scan cycle; the first time encoding cycle is denoted as... s), the second time coding period is denoted as (s), the least common multiple of the two periods is denoted as (s, defined as) The first time-encoded phase offset is denoted as (rad), the second time-coded phase offset is denoted as (rad), the time resolution period of the event image acquisition unit is denoted as (s). The triggering and phase scheduling submodule is used in each The internal repeat output encoding and trigger sequence preferably satisfy the following non-aliasing condition: , The aforementioned phase reversal and sampling lower bound conditions are used to improve the separability of events in two channels within the same period.

[0068] The time encoding and gating generation submodule is used to generate the gating function and trigger clock for two channels; the half-width of the gating window is denoted as... (s), and with , The time interval for receiving valid events is jointly defined. For ease of explanation, the first channel gating function is defined. With the second channel gate function (dimensionless), at any time (s) satisfies:

[0069]

[0070]

[0071] in Represents the modulo operation; when Only the event of the first bandpass tag is received. Only events from the second bandpass tag are received; the time encoding and gating generation submodule is used to ensure... and , Relationship satisfaction This reduces the probability of window overlap.

[0072] The jitter suppression and phase servo submodule is used to perform closed-loop correction of the phase offset; the estimated deviation of the cross-channel arrival time is denoted as... (s), given by the processor based on cross-correlation calculation; the proportional coefficient of the phase servo is denoted as (Dimensionless), the integral coefficient is denoted as (Dimensionless), jitter suppression and phase servo submodules are used for pressing , Perform bidirectional fine-tuning to maintain The target relationship is to suppress short-term phase jitter.

[0073] The frequency drift compensation submodule is used to perform frequency-level correction on the time coding period and the peripheral sampling period; where the first time coding frequency is denoted as... (Hz), the second time-coded frequency is denoted as (Hz), the sampling frequency of the event acquisition link is denoted as (Hz), the frequency error estimate is denoted as... , , (Hz); The frequency drift compensation submodule is used to calculate the frequency drift according to the frequency. , , Low-amplitude online correction is performed, and the correction amount is synchronously written into the local timing generator of the dual-cone projection unit and the event image acquisition unit.

[0074] The health monitoring and watchdog submodule is used to continuously evaluate time-spectral consistency and safety boundaries during operation; the gating overlap rate of the two channels is denoted as... (Dimensionless, defined as a unit of time) and (Time percentage), event congestion indication quantity is recorded as (Dimensionless, derived from the buffer occupancy rate of the event image acquisition unit), the phase deviation statistic is denoted as... (rad, defined as) The standard deviation), channel crosstalk proxy quantity is denoted as (Dimensionless, estimated by the rate of out-of-bounds events within the window). The health monitoring and watchdog submodule is used when... Exceeding the threshold or Exceeding the threshold or Exceeding the threshold When this happens, three protection mechanisms are automatically triggered: the first is to temporarily narrow the door-controlled window by half a width. Secondly, it increases the phase offset difference in a short time. The perturbation amount is used to open the activity window; thirdly, the frequency reduction and load reduction (reduce) , This helps alleviate congestion and writes events to the log.

[0075] The register and event log submodule is used to read and write parameters and status with the processor; the control register group is used to receive... , , , , , , , With protection threshold The status register group is used for reporting. , , , , The event log records the timestamp correction amount, frequency correction amount, and out-of-bounds reason, and shares it with the result generation unit for traceability, in relation to the most recent protection action.

[0076] Regarding its coordination with other units in the system, the synchronization control unit is used to: coordinate with the event image acquisition unit in terms of time resolution period and gating function. , A consistent reception rule is formed; a consistent projection rule is formed with the biconical projection unit in terms of time coding period and phase offset; spectral-temporal consistency is maintained with the band separation unit in terms of bandpass switching timing; and closed-loop switching with the processor is achieved. It works with quality metrics to drive phase servoing and frequency compensation; collaborates with the cone parameter solving unit to improve the condition number of cone parameter estimation by using gated events with low overlap and low jitter; collaborates with the scale self-calibration unit to improve the observability of intersection geometry through stable period and phase relationships; and shares health quantification indicators with the results generation unit to form traceable records.

[0077] Regarding connectivity and interaction, the synchronization control unit receives data through: the control register set. With protection threshold set; status register group reporting The log interface outputs time correction, frequency correction, and out-of-bounds events to the result generation unit, along with the most recent action.

[0078] Through the above structure and process, the synchronous control unit provides a stable, adjustable and traceable time skeleton for simultaneous encoding of two channels under the combination of "master time base synchronization + phase scheduling + gating generation + phase servo + frequency compensation + health monitoring", and uses quantified overlap rate, phase deviation and misclassification proxy quantity as health indicators.

[0079] This embodiment will describe the cone parameter solving unit in detail. The cone parameter solving unit includes: a data access and preprocessing submodule, a pixel-ray mapping submodule, a biconical joint parameterization submodule, an initial value generation submodule, a robust least squares solving submodule, an observability assessment and constraint management submodule, an online consistency and quality assessment submodule, and a parameter publishing and caching submodule. The cone parameter solving unit resides in the processor and exchanges intermediate quantities with the storage unit. The cone parameter solving unit receives a first event set and a second event set from the event image acquisition unit through a first data interface, receives pixel-level bandpass tags from the band separation unit through a second data interface, receives time coding and gating functions from the synchronization control unit through a third data interface, and receives design parameter prompts (including the opening angle setting range and nominal vertex spacing value) from the biconical projection unit through a fourth data interface. It then publishes the estimated opening angle, vertex position, and axial direction parameters to the scale self-calibration unit and the depth reconstruction unit.

[0080] Data access and preprocessing:

[0081] The data access and preprocessing submodule is used to filter and deduplicate events within the same scan cycle according to band labels and gating functions, obtaining a sparse pixel set for geometric solution; where the pixel coordinate set of the first event set is denoted as... The set of pixel coordinates of the second event set is denoted as Each pixel coordinate is represented in homogeneous form. , (Unit: pixels); Gating function , Events are retained if they match the channel label; otherwise, they are discarded. To suppress cluster redundancy, the data access and preprocessing submodules perform non-maximum suppression on multiple events within the same pixel neighborhood. The side length of the non-maximum suppression window is denoted as... (Unit: pixels).

[0082] The pixel-to-ray mapping submodule is used to map pixel coordinates to unit line-of-sight vectors in the camera coordinate system; where the camera intrinsic matrix is ​​denoted as... The camera's optical center coordinates are as follows (m), pixel The normalized line-of-sight vector is denoted as (Dimensionless unit vector), pixel The normalized line-of-sight vector is denoted as Any three-dimensional point falling within the line of sight can be represented as , ,in is the depth parameter (m).

[0083] Biconical joint parameterization:

[0084] The dual-cone joint parameterization submodule is used to analytically describe two cones using a shared axis; where the vertex position of the first cone is denoted as... (m), the position of the vertex of the second cone is denoted as (m), the unit vector along the common axis is denoted as (Dimensionless), the opening angle of the first cone surface is denoted as... (rad or angle), the opening angle of the second cone is denoted as (rad or angle), the known distance between the vertices of the two cones along the projection optical axis is denoted as (m). For any three-dimensional point Define "axial projection" and "radial projection":

[0085] , ,

[0086] in Represents a third-order identity matrix. The conical surface satisfies the implicit equation... .

[0087] Therefore, the residual functions of the first and second conical surfaces are defined as follows: , .in This represents the set of parameters to be estimated. To reflect physical consistency, the biconical joint parameterization submodule... Two soft constraints are applied: vertex coaxiality-spacing constraint and angular difference lower bound constraint. , ,in Set the lower limit value (rad or angle) for the design.

[0088] Initial value generation: The initial value generation submodule is used to provide convergent initial parameter values ​​in the absence of an external calibration board. This submodule consists of two steps:

[0089] Initial value of the axial direction: Assume that the initial value of the common axial unit vector is parallel to the camera's optical axis, and denote the initial value of the common axial unit vector as... ; through the ; The event density map is used to perform polar coordinate voting to obtain a coarse relationship between the principal radius of the ring and the depth. , The rough estimate, the initial value of the opening angle is denoted as follows: , .

[0090] Initial values ​​of the vertices: Let the initial values ​​of the vertices of the second cone be [values ​​to be filled in The initial value of the first cone vertex Taken on the optical axis of the camera, at a distance from the optical center of the camera. Location, By the principal radius of the ring and Consistent coarse registration is obtained. The initial value set is denoted as... .

[0091] Robust Least Squares Solution: The Robust Least Squares Solution submodule is used for joint optimization under pixel-ray constraints. For each pixel ray parameter can be or Analytical (or numerical) elimination yields the random variable. The smallest positive root of the change Subsequently, using a strategy of one-layer elimination and outer-layer parameter optimization, the objective of the following expression is minimized: ,in Represents the robust cost function (preferably in Huber form). , The constraint weights are set. To improve convergence speed, the robust least squares solution submodule adopts a Gaussian-Newton and Levenberg-Marquardt switching strategy to automatically select the trust region radius; simultaneously, the Jacobian matrix is ​​calculated. condition number ,when Exceeding the threshold When this occurs, the observability assessment and constraint management submodule is triggered to perform parameter rescaling and weight reallocation.

[0092] Observability Assessment and Constraint Management: The Observability Assessment and Constraint Management submodule is used to determine whether the current data is sufficient for stable estimation. And, when necessary, collaborate with other units to improve conditions. First, define the intersection visibility index. (Dimensionless, defined as the proportion of pixels that are illuminated by both conical surfaces and successfully indexed within the same scan cycle), when or At that time, the observability assessment and constraint management submodule is used to suggest to the synchronization control unit to temporarily narrow the gate window half-width or fine-tune the phase offset, and to suggest to the biconical projection unit to fine-tune the drive frequency difference to change the event timing pattern; within the optimization, robust weights are increased for edge sectors and low-density regions, and the sector weight gain is denoted as... (Dimensionless) to suppress the bias caused by uneven sample distribution.

[0093] Online Consistency and Quality Assessment: The online consistency and quality assessment submodule is used to provide parameter quality quantification after convergence; where the standard deviation of the opening angle is denoted as... , (rad or angle), the standard deviation of the vertex position is denoted as . , (mm), axial direction uncertainty is denoted as (rad), vertex spacing consistency error is denoted as (mm), the median residual is denoted as (In terms of the dimensions of the equation residuals). When or When the set threshold is exceeded, the online consistency and quality assessment submodule is used to trigger reinitialization or request the processor to perform a short-term back sampling.

[0094] Parameter publishing and caching: The parameter publishing and caching submodule is used to publish the final parameters. and quality The data is distributed to scale self-calibration units and depth reconstruction units, and the most recently used data is stored in a circular cache. The estimation results for each cycle are used for traceability by the result generation unit. The ring cache capacity is denoted as... (Dimensionless counting).

[0095] Single-ray resolution depth: For a given pixel ray and a given cone, substitute... Afterwards, one can obtain information about The quadratic equation: Its physical root is taken as the smallest positive root. .

[0096] Jacobi closed expression: targeting , , The partial derivatives can be obtained from the above equation using the chain rule; to ensure Unit range constraints, introducing local biparameterization (e.g., biangular parameterization in spherical coordinates) or adjusting after each iteration Normalization.

[0097] Scale Consistency Interface: To cooperate with the scale self-calibration unit, the cone parameter solving unit provides the set of image plane intersection points of the two cone surfaces and the straight line features of the wall surface, and the corresponding... So that only one can be used later. Complete the analytical solution for the pixel-length scaling factor.

[0098] Connection and interaction: The cone surface parameter solving unit receives data via a sampling interface. Gating functions and bandpass tags; indicating the aperture angle range and nominal spacing of the biconical projection unit received by the interface. The control interface obtains the set of optimized hyperparameters and observability thresholds from the processor. The publishing interface outputs the final parameters and sparse intersection points to the scale self-calibration unit and the depth reconstruction unit; the log interface outputs to the result generation unit. The time series is used for tracing the source.

[0099] Coordination with other system units: The cone parameter solving unit is used for: consistently selecting valid events with the event image acquisition unit under gating and timing; consistently dividing the two event sets on pixel-level bandpass labels with the band separation unit; coordinating with the synchronization control unit on phase and window width to improve observability; sharing nominal geometry with the biconical projection unit to accelerate initial value generation; and supporting the scale self-calibration unit by providing a set of image plane intersection points of the two cones and wall straight-line features using only... Absolute scale resolution; sharing per pixel with depth reconstruction units. The cost of conical surface consistency is used to implement pixel-level fusion; quality quantification metrics are shared with the result generation unit to form a traceable record.

[0100] Through the above structure and process, the cone parameter solving unit, under the combination of "shared axis joint parameterization + single-ray analytical elimination + robust least squares + observability closure", can stably, interpretably and quantifiably obtain the opening angle, vertex position and axial direction of the two cones in a single scan cycle, and provide high-quality priors for the scale self-calibration unit and the depth reconstruction unit.

[0101] This embodiment will describe the scale self-calibration unit in detail. The scale self-calibration unit includes: an intersection extraction submodule, a double-loop intersection point calculation submodule, an axial projection line estimation submodule, a pixel-length scaling factor calculation submodule, an online smoothing and confidence assessment submodule, an anomaly detection and rollback submodule, and a parameter publishing and archiving submodule. The scale self-calibration unit resides in the processor and exchanges intermediate quantities with the storage unit. The scale self-calibration unit receives the opening angle and geometric description of the two cones from the cone parameter solving unit, the known distance between the vertices of the two cones along the projection optical axis from the double-cone projection unit, and the event point set and pixel-level bandpass labels of the two channels from the event image acquisition unit and the band separation unit. It then publishes the obtained pixel-length scaling factor and quality quantity to the depth reconstruction unit and the result generation unit.

[0102] The symbols appearing in this embodiment are defined as follows: the known distance between the vertices of the two conical surfaces along the projection optical axis is denoted as... (m); the opening angle of the first cone surface is denoted as (rad or angle), the opening angle of the second cone is denoted as (rad or angle); the pixel-length scaling factor to be determined is denoted as... (m / pixel, used to convert pixel distance in the image to physical length); the fitted curve of the intersection of the first cone and the wall on the image plane and the fitted curve corresponding to the second cone are respectively denoted as... , (A quadratic curve obtained by fitting the event set using the cone surface parameter solution unit); a straight line feature of a wall surface belonging to the same category as the wall surface on the image plane is denoted as... The geometric centers (image projection centers) of the two fitted curves are denoted as follows in the pixel coordinate system: and (Unit: pixels), thus defining the axial projection line as the connection and Let A be the denoted .

[0103] Intersection extraction and intersection point determination: The intersection extraction submodule is used to robustly extract at least one wall line feature from the event density map on the image plane. The straight line features of the wall surface are preferably derived from the wall-to-ground intersection line, the shadow hard edge, or the decoration seam line. Probabilistic Hough or extreme geometric consistency is used for robust fitting, and the straight line equation and its confidence level are output.

[0104] The submodule for finding the intersection of two loops on the same line is used to calculate... and respectively with The set of intersection points; let the directed distance (relative to) the intersection point with the first cone on the image plane be... (Positive along the projection direction of A) (Unit: pixels), the intersection point with the second cone surface relative to The corresponding directed distance is (Unit: pixels). To improve robustness, in and , Take a set of intersection points on each side to obtain (index =1,2... represent sample numbers), and define the double-ring difference along the same line. (Unit: pixels).

[0105] Axial projection line estimation and alignment: The axial projection line estimation submodule is used to estimate the projection of the common axis of the two conical surfaces onto the image plane. Through connection and After obtaining the initial value, fine-tuning is performed using the criterion of minimizing the radial asymmetry between the two curves in that direction; once completed, Projected onto the local coordinate system Above, to ensure , The measurement direction is consistent with the radius direction of the cone, thereby reducing the impact of perspective distortion on scale calculation.

[0106] Analytical solution for pixel-length scaling factor (single-point closed-form + multi-point robust): The pixel-length scaling factor solver module is used to provide an analytical solution based on coaxial double-loop geometry without relying on external calibration plates or camera intrinsic parameters. The analytical solution and robust estimation are derived. The theoretical basis is: on the axial projection line, the first cone surface is at a distance... The radius at (m) is The second cone surface is on the same The radius at that point is The directed pixel distance on the image plane is linearly related to the physical radius by the pixel-length scaling factor. , This leads to the co-linear double-loop difference. Rearranging the above equation, we can directly obtain the closed-form solution for the pixel-length scaling factor: The numerator and denominator have dimensions of meters and pixels, respectively. The dimension of is meters per pixel. The above formula applies to any set of... Established (in) Sampling from the same side and with consistent signs is more robust, therefore each sample can be sampled... Calculate one .

[0107] To improve noise resistance, the pixel-length scaling factor solution module employs multi-point robust estimation: for each sample calculate Outliers exceeding three times the absolute deviation of the median were removed, and the weighted median of the remaining samples was used as the median. The point estimates are weighted proportionally to the event density at that sample and the reciprocal of the curve fitting residuals. This process does not require a solution. It avoids dependence on the pose of the camera and conforms to the constraint of "using only known spacing, opening angle and geometric quantities of intersection points of the same line".

[0108] Online Smoothing and Confidence Assessment: The online smoothing and confidence assessment submodule is used to fuse the point estimate of the current period with historical values ​​and output the uncertainty. The scaling factor for the previous period is defined as... (m / pixel), the point estimate for the current period is (m / pixel), online weight is denoted as (dimensionless, derived from event density, , (Given by a combination of the fitting residual and the sample size), then .

[0109] Simultaneously output the confidence interval: denoted as the within-sample variance. ( / ), then give The interval is used as the operating uncertainty; and the scale drift indicator is calculated. (Dimensionless) is used to monitor temperature drift or optical path disturbances.

[0110] Anomaly detection and rollback:

[0111] The anomaly detection and rollback submodule is used to trigger protection in the following situations:

[0112] Geometric degradation of two loops along the same line: when Too small or two fitted curves and When the geometric angle is less than the threshold, it is marked as a degraded frame;

[0113] Insufficient events: When the number of samples on the same side is less than the set lower limit or the event density is lower than the threshold, it is marked as an insufficient frame;

[0114] Inconsistency violation: when or When the limit is exceeded, it is marked as an inconsistent frame.

[0115] If any of the above conditions are met, the rollback will revert to the previous reliable result. It requests the processor to drive the synchronization control unit to briefly expand the gate window or drive the dual-cone projection unit to fine-tune the frequency difference to increase the sample density; if necessary, it triggers the shrinkage of the region of interest of the event image acquisition unit to focus on the vicinity of the wall-ground intersection line and quickly restore observability.

[0116] Parameter publishing and archiving: The parameter publishing and archiving submodule is used to publish and archive parameters. and its uncertainty Published to the deep reconstruction unit, and The time series data is written to the result generation unit and storage unit, while saving key intermediate values. (Sample index and reasons for removal) to facilitate tracing the source.

[0117] It needs to be explained that:

[0118] Directional consistency: To avoid sign reversal, The direction is based on the positive direction of A; when sampling from both sides, the same side sign convention is maintained.

[0119] Noise robustness: when , When obtaining the value from an ellipse, first perform radial linearization along A, retaining only the intercept and slope in that direction to achieve a consistent correspondence with the cone radius.

[0120] Numerical stability: when and When the sample sizes are close, increase the sample size and use a weighted median to avoid the denominator being too close. "The amplification effect caused by being too small."

[0121] Coordination with other system units: The scale self-calibration unit is used to: share the analytical description of the two cone surfaces on the image plane with the cone parameter solving unit and obtain... , It acquires two-channel events within the same scanning cycle as the event image acquisition unit and provides pixel-level bandpass tags by the band separation unit; it requests temporary expansion of the gating window or fine-tuning of the phase with the synchronization control unit when needed to increase sample density; it shares the final pixel-length scaling factor and its uncertainty with the depth reconstruction unit to establish an absolute scale; and it shares the scale time series and intermediate quantities with the result generation unit to form a traceable record.

[0122] Through the above structure and process, the scale self-calibration unit can stably obtain the pixel-length scaling factor without external calibration and camera intrinsic parameters under the combination of "coaxial double-loop intersection geometry + axial projection alignment + closed analytical solution + multi-point robustness + online smoothing", and form a closed loop with uncertainty and drift indication as quality indicators.

[0123] See Figure 4 As shown, the depth reconstruction unit will be described in detail in this embodiment. The depth reconstruction unit includes: a data aggregation submodule, a pixel-ray construction submodule, a cone intersection submodule, a pixel-level consistency and threshold determination submodule, a weighted fusion and invalid mask generation submodule, a densification and hole repair submodule, a coordinate transformation and surface regularization submodule, and a quality assessment and logging submodule. The depth reconstruction unit resides in the processor and exchanges intermediate quantities with the storage unit. The depth reconstruction unit receives pixel events within the same scanning cycle from the event image acquisition unit, receives pixel-level bandpass tags from the band separation unit, receives gating functions from the synchronization control unit, receives parameters such as the opening angle, vertex position, and axial direction of the two cone surfaces from the cone parameter solution unit, and receives the pixel-length scaling factor and its uncertainty from the scale self-calibration unit. It also publishes the fusion height, effective pixel mask, and quality quantity to the flatness evaluation unit, the indoor coordinate establishment unit, and the result generation unit.

[0124] It should be noted that, for clarity and consistency, the symbols used in this embodiment will be explained here:

[0125] The first event set and the pixel coordinate set are denoted as... (Pixel coordinates in homogeneous form) (represented); the second event set, the pixel coordinate set, is denoted as ;

[0126] The camera intrinsic parameter matrix is ​​denoted as The three-dimensional coordinates of the camera's optical center are as follows: (m);

[0127] The position of the vertex of the first cone is denoted as (m), the position of the vertex of the second cone is denoted as (m), the unit vector along the common axis is denoted as (Dimensionless), the opening angles of the first and second conical surfaces are denoted as... , (rad or angle), the known distance between the vertices of the two cones along the projection optical axis is denoted as (m);

[0128] The pixel-to-length scaling factor is denoted as (m / pixel, used to convert pixel distance to physical length), its uncertainty is denoted as (m / pixel);

[0129] Pixels The height estimates obtained from the first and second cone surfaces are denoted as follows: , (m), pixel blending height is denoted as m);

[0130] The adaptive consistency threshold function is denoted as (m), the normalized cone surface cost is denoted as (m) and (m) respectively. , (Dimensionless), the upper limit of the cost setting is denoted as (dimensionless);

[0131] Pixel weight is denoted as , (dimensionless, satisfies) );

[0132] This represents the variable that minimizes the objective function.

[0133] Data aggregation and pixel-ray construction:

[0134] The data aggregation submodule is used to deduplicate and register events from two channels within the same scan cycle, and constructs a candidate set indexed by pixels. For each pixel coordinate... or The pixel-ray construction submodule is used to generate unit gaze vectors based on camera intrinsics: , ,in Let be a dimensionless unit vector. Any three-dimensional point along a line of sight can be written as . , ,in is the depth parameter (m).

[0135] Finding the intersection of cones (generating the heights of two paths):

[0136] The cone intersection submodule is used to calculate the intersection points of the pixel ray with the two cone surfaces and obtain two height estimates. To define the implicit equations of the cone surfaces, let the axial projection and radial projection be... , ,in It is a third-order identity matrix. The conical surface satisfies... .Will or Substituting the values, we can obtain information about... Find the smallest positive root of the quadratic equation. This serves as the depth of the physical intersection point. Furthermore, the height of the two paths at a pixel is defined as... , ,in , These are the minimum positive roots of the pixel ray and the first and second cone surfaces, respectively. For ease of quality measurement, the normalized cone surface cost is defined.

[0137]

[0138]

[0139] The denominator above is a scaling term, used to suppress dimensional amplification at long distances.

[0140] It should be noted that when the camera intrinsics are unavailable or not yet calibrated, the conic intersection submodule can degenerate into an axial projection approximation path: along the axial projection line of the image plane using a pixel-length scaling factor. Convert the directional distance of pixels into a physical radius, and use... , Solving closed relations , The path is geometrically consistent with the scale self-calibration unit.

[0141] Pixel-level consistency and threshold determination:

[0142] The pixel-level consistency and threshold determination submodule is used to calculate an adaptive threshold based on the height and geometric differences between the two paths and to filter valid pixels. The adaptive threshold function is defined as follows: ,in This represents the basic threshold setting value (m). This represents the proportionality coefficient setting value (dimensionless). When... and , When the pixel is in the correct state, it is marked as consistent and enters the fusion process; otherwise, it is marked as invalid and written into the invalid mask.

[0143] Weighted fusion and invalid mask generation:

[0144] The weighted fusion and invalid mask generation submodule is used to output the fused height at the pixel level. Pixel weights are jointly given by event quality and spectral quality. , ,in , The uncertainty is determined by the local event density of the event image acquisition unit, the band communication-to-noise ratio of the band separation unit, and the scale self-calibration unit. The mass fraction (dimensionless) constituted by the combined components. The fusion height adopts a first-order robustness criterion: .

[0145] The closed-form solution to the above equation is the weighted median: when and and When the minimum point is not unique, the value closer to the one with the larger weight is taken to stabilize the value. The resulting output is a binary invalid mask. (dimensionless, (Indicates an invalid pixel).

[0146] Denseization and Void Repair:

[0147] The Denseening and Hole Repair submodule is used to fill in invalid regions while preserving geometric boundaries. First, edge-preserving guided interpolation is performed, with the neighborhood window radius denoted as... (Unit: pixels), the guiding weight is given jointly by event density and gradient; then a topology-preserving morphological closing operation is performed to remove isolated holes; for large invalid regions (area threshold denoted as...), ... The unit is Maintain voids to avoid introducing false surfaces. This submodule outputs a dense heightmap. With updated mask .

[0148] Coordinate transformation and surface regularization:

[0149] The coordinate transformation and surface regularization submodule transforms the dense height map from the camera coordinate system to the indoor coordinate system and performs lightweight regularization. The transformation matrix is ​​provided by the indoor coordinate establishment unit, and surface regularization is performed after aligning the wall normal with the gravity direction: objective function ,in The height after regularization. This is the set of valid pixels (consistent with M). These are the first-order regularization weights (dimensionless). The regularized output is a surface height map used for subsequent fitting and evaluation.

[0150] Quality assessment and logs:

[0151] The quality assessment and logging submodule is used to record key quality metrics: pixel-level consistency rate. (Dimensionless) Statistics of fused residuals (m), void ratio, regularization loss value, etc., and share the time series with the result generation unit for traceability.

[0152] It should be noted that the key points of verification and implementation in this implementation are as follows:

[0153] Stable solution to quadratic equations: finding the quadratic expression through the intersection of cones. A numerically stable root-finding strategy is adopted. To reduce the elimination error, only the smallest positive root is taken as the physical solution.

[0154] Threshold-adaptive scale coupling: when When increasing, appropriately relax the restrictions. upper limit ( (These are dimensionless coefficients) to avoid false rejections due to scale uncertainty.

[0155] Mask propagation: The pixels propagate in a confidence-attenuated manner within the neighborhood, reducing the impact of isolated outliers on the fit.

[0156] Parallelization: Cone intersection and fusion are independent at the pixel level, and linear acceleration can be achieved by using data parallel threads and vector instructions.

[0157] Coordination with other system units: The depth reconstruction unit is used for: consistently converging two-channel events with the event image acquisition unit under time synchronization and gating strategies; sharing pixel-level bandpass labels with the band separation unit to ensure channel differentiation; sharing conical geometry with the cone parameter solving unit for intersection calculation; sharing pixel-length scaling factor and uncertainty with the scale self-calibration unit for adaptive thresholding; collaborating with the synchronization control unit to adjust phase and window width to restore observability under abnormal conditions; sharing coordinate alignment parameters with the indoor coordinate establishment unit to output a height map in the indoor coordinate system; sharing fused height, effective mask, and quality quantity with the flatness evaluation unit to complete the evaluation; and sharing key logs with the results generation unit to form a traceable record.

[0158] Through the above structure and process, the deep reconstruction unit, under the combination of "cone intersection + pixel-level consistency threshold + robust weighted fusion + densification + regularization", can stably recover the wall density height map from two-channel events, and use quantifiable residuals, void ratio and consistency rate as quality indicators.

[0159] See Figure 5As shown in this embodiment, the flatness evaluation unit will be described in detail. The flatness evaluation unit includes: an input integration submodule, an evaluation mesh generation submodule, a weighted plane fitting submodule, a flatness index calculation submodule, an anomaly and hole handling submodule, a multi-scale summary and compliance mapping submodule, and a quality assessment and recording submodule. The flatness evaluation unit resides on the processor in software and exchanges intermediate quantities with the storage unit. The flatness evaluation unit receives the fused height map and effective pixel mask from the depth reconstruction unit, receives coordinate alignment parameters from the indoor coordinate establishment unit, receives the opening angle and vertex spacing hints of the two cones from the cone parameter solving unit, receives the pixel-length scaling factor and its uncertainty from the scale self-calibration unit, and publishes the mesh-level and global-level results to the result generation unit for overlay and archiving.

[0160] It should be noted that, for clarity and consistency, the symbols used in this embodiment will be explained here:

[0161] The fusion height map is noted as (m), whose pixel domain definition set is denoted as (Set of pixel indices, dimensionless); the effective pixel mask is denoted as... (dimensionless) 1 indicates that it is valid. (Indicates invalid)

[0162] The planar coordinates of pixel coordinates in the indoor coordinate system are denoted as (m), corresponding to the height is (m);

[0163] The opening angles of the two cones are denoted as follows: , (rad or angle), the known distance between the vertices of the two cones along the projection optical axis is denoted as (m);

[0164] The side length of the evaluation grid is denoted as (m), determined by the following formula ,in This represents an interval-limiting function (which restricts real numbers to a given upper and lower limit). This represents the grid scale factor setting value (dimensionless).

[0165] The pixel-level weighting function is denoted as (Dimensionless), used for plane fitting and flatness index calculation; to be consistent with the aforementioned system definition, take... ,in , These are the two heights (m) given by the depth reconstruction unit. Let m represent the adaptive consistency threshold. and Set value;

[0166] Weighted flatness index is marked as (m) is the main index of the grid-level output.

[0167] Evaluation of mesh generation:

[0168] The evaluation mesh generation submodule is used to construct a square evaluation mesh on the wall area in the indoor coordinate system, with the mesh side length defined by the above formula. The wall boundary is determined by the region of interest on site or the full-frame image; the grid index set is denoted as... (Dimensionless counting), each grid The set of valid pixels covered is denoted as Incomplete grid edges are detected when the number of samples is less than a threshold. When performing dimensionless counting, the index is not calculated temporarily to ensure statistical stability; if necessary, adaptive expansion to adjacent grids is used to supplement pixels.

[0169] Weighted plane fitting:

[0170] The weighted plane fitting submodule is used to perform plane fitting on local wall surfaces within each evaluation grid. The local plane model is written as... ,in , , These are the plane coefficients (units are dimensionless, dimensionless, and meter, respectively). Let the grid be... The sample matrix and vector are: , , ,

[0171] The weighted least squares solution is: .

[0172] To improve numerical stability, the weighted plane fitting submodule is used when the matrix condition number exceeds a threshold. Perform column standardization and eigenvalue truncation as needed, and adjust accordingly. , Decentralization is performed (with the grid center as the zero point) to reduce correlation.

[0173] Calculation of flatness index:

[0174] The flatness index calculation submodule is used to output various flatness-related quantities within each evaluation grid, with the main index being the weighted flatness index. (m). Simultaneously output auxiliary indicators:

[0175] Unweighted maximum residual Root mean square residual Coverage (dimensionless, The average pixel area (per pixel), used to quantify the sufficiency of valid samples within a grid. These multiple indicators are written together into the result generation unit along with the grid number, facilitating subsequent visualization and compliance comparison.

[0176] Anomaly and Void Handling:

[0177] The anomaly and hole handling submodule is used to correct or mark the mesh in the following situations:

[0178] Insufficient sample size: when At this time, no value will be output, and it will only be marked as "Insufficient data";

[0179] Abnormally high residuals: when Exceeding the threshold And coverage When sufficient, trigger local robust refitting (Huber or quantile regression) to reduce the impact of isolated outliers;

[0180] Large area voids: When the area where M is zero within the grid exceeds a threshold proportion. At that time, the void state is preserved and a rescan is prompted (the processor coordinates the synchronous control unit and the event image acquisition unit to perform a short-term rescan).

[0181] Multi-scale aggregation and compliance mapping: The multi-scale aggregation and compliance mapping submodule is used to verify the smoothness results at different grid scales and form a global summary. Define the scale sequence. , , Calculate the main and auxiliary indices at each scale, and output the maximum main index across the multiple scales. (m) is used to address the unevenness of different spatial frequencies. If it is necessary to compare with the engineering acceptance clauses, the multi-scale summary and compliance mapping submodule can map the main indicators to three-level labels of "qualified / warning / unqualified" (the threshold is given by the project configuration) and output them to the result generation unit along with the basis.

[0182] Quality assessment and recording:

[0183] The Quality Assessment and Recording submodule is used to record key quality quantities at the grid and global levels, including: matrix condition number. The system generates statistics, coverage histograms, quantiles of main indicator distributions, hole percentages, and robust refit trigger counts; and shares time series and spatial distribution heatmaps with the results generation unit to form a traceable record. When necessary, the parameter groups for each grid are... Output along with the overlay layer, where (m), , The minimum and maximum values ​​(m) of the adaptive threshold within the grid. This is the duty cycle vector for dual-channel events (dimensionless).

[0184] Coordination with other units in the system:

[0185] The flatness evaluation unit is used to: share fusion height and invalid masks with the depth reconstruction unit to construct samples; share alignment parameters with the indoor coordinate establishment unit to ensure planar semantics; share opening angle, vertex spacing and pixel-length scaling factor with the cone parameter solving unit and the scale self-calibration unit to set the grid scale and adaptive threshold; share grid-level and global-level indicators, coverage and quality with the result generation unit and overlay them onto the wall image to form a report; and, when necessary, coordinate with the synchronization control unit and the event image acquisition unit to perform back-acquisition to improve the gap data.

[0186] Through the above structure and process, the flatness evaluation unit, under the combination of "adaptive mesh + weighted plane fitting + main / auxiliary multi-index + multi-scale verification + anomaly and void handling", can stably output wall flatness evaluation results with engineering usability in a well-aligned indoor coordinate system.

[0187] This embodiment will describe the result generation unit in detail. The result generation unit includes: an input integration submodule, a coordinate and layer construction submodule, a visualization mapping and overlay submodule, a statistical index summary and classification submodule, a report generation and export submodule, an audit and traceability submodule, and a runtime display and slice publishing submodule. The result generation unit resides on the processor in software and exchanges intermediate quantities with the storage unit. The result generation unit receives grid-level and global-level flatness indices from the flatness evaluation unit, dense height maps and invalid masks from the depth reconstruction unit, indoor coordinate transformation parameters from the indoor coordinate establishment unit, and the opening angle, vertex spacing, and pixel-length scaling factor and their uncertainties of the two cone surfaces from the cone parameter solving unit and the scale self-calibration unit. It then outputs the generated overlay layer, statistical data file, and report document to the storage unit and provides them for on-site display.

[0188] It should be noted that, for clarity and consistency, the symbols used in this embodiment will be explained here:

[0189] The dense height map is denoted as Z (m), and the effective pixel mask is denoted as... (dimensionless);

[0190] The indoor coordinate plane position of each pixel is denoted as... (m), height is (m);

[0191] The evaluation grid set is denoted as (Dimensionless counting), the sample set for each grid is denoted as (Set of pixel indices, dimensionless);

[0192] The grid-level main indicator (weighted flatness indicator) is denoted as: (m), the unweighted maximum residual is denoted as (m), the root mean square residual is denoted as m), coverage is denoted as (dimensionless);

[0193] The quantile of the global aggregate metric is denoted as (m, where) (representing quantiles), the maximum value is denoted as (m), the median is denoted as (m);

[0194] The color mapping function is denoted as (Dimensionless, input is a real number, output is a ternary color vector), the transparency mapping function is denoted as... (dimensionless);

[0195] The background image pixel color is denoted as (Dimensionless ternary vector), the superimposed pixel color is denoted as (Dimensionless ternary vector), the output pixel color is denoted as (Dimensionless ternary vector).

[0196] Coordinates and Layer Construction:

[0197] The Coordinates and Layers building submodule is used to organize height and mesh results into overlayable layer objects. This submodule creates three types of layers:

[0198] Measurement layer: Records the residual measurement or height deviation of each pixel at the pixel level; the pixel measurement is denoted as... (m), by default, takes the absolute deviation of the local plane. .

[0199] Mesh layer: Stores each mesh as a polygon vector. The boundary and its indicator set .

[0200] Annotation layer: Stores directional arrows, scale bars, text annotations, and compliance classification labels. The geometry of all layers is expressed in the coordinate system given by the indoor coordinate establishment unit, ensuring that the position is consistent with the actual measured space; the mapping from pixels to indoor coordinates is provided by known projection relationships and pixel-length scaling factors.

[0201] Visual mapping and overlay:

[0202] The visualization mapping and overlay submodule is used to map metrics to colors and then transparently overlay them onto a background image. For robust visualization, the upper and lower bounds of the visualization range of the metrics are first calculated:

[0203] , ,

[0204] in and The truncation quantiles are set to a dimensionless value, typically 0.02 or 0.98, to suppress extreme outliers. The normalization value is then defined. Color mapping and transparency mapping are respectively , ,in Transparency benchmark (dimensionless). The transparency gamma factor is dimensionless. Overlay uses standard foreground-background linear blending. .when When the value is 0, it is displayed as a hollow area using a grid pattern or a low-saturation grayscale, facilitating the identification of missing measurement areas on-site. The grid boundary and text annotations are overlaid with vectors to ensure no distortion during scaling.

[0205] Summary and classification of statistical indicators:

[0206] The statistical indicator summary and hierarchical submodule is used to generate global statistics from grid sets and pixel metrics. This includes the main indicator set. calculate , , And simultaneously calculate the coverage-weighted root mean square index. .

[0207] To connect with project acceptance thresholds, define an upper limit threshold. With warning threshold (m, from project configuration), generate three-level tags using the following formula:

[0208]

[0209] Labels are marked with both color and text on a grid layer for easy and quick verification.

[0210] Report generation and export:

[0211] The report generation and export submodule is used to organize visualization results and statistical tables into normalized output, including:

[0212] Image overlay files: Output in vector-raster hybrid format (PDF / SVG / PNG), including metric heatmaps, grid boundaries, direction arrows, scale bars, and key statistics;

[0213] Statistical data files: Output in structured text and tabular formats (CSV / JSON); JSON entries include grid number, center coordinates, Grading labels and necessary quality quantities;

[0214] Multi-page report document: Automatically generates overview page (global heatmap and global statistics), detail page (indicators listed by grid and local magnified view), and quality page (coverage histogram, matrix condition number statistics, hole percentage and number of triggered refits).

[0215] Before exporting, all numerical units should be uniformly declared in the header; the legend should include color-to-millimeter reference scales and be labeled. , With quantile position.

[0216] Audit and traceability:

[0217] The auditing and traceability submodule ensures the integrity and traceability of the results. First, it generates fingerprint digests for key output files: (File byte sequence), where H is a hexadecimal string (dimensionless). The generation context is also recorded, including: the opening angle of the two cones, vertex spacing, pixel-to-length scaling factor and uncertainty, gating parameters, version number, and timestamp. {H, context} is written to the audit log and embedded in a QR code at the end of the report for easy external verification. To enable version replay, a snapshot of the configuration and color-coded parameters used in the generation are saved, forming a one-to-one correspondence between "result" and "configuration".

[0218] Runtime Display and Tile Publishing: The Runtime Display and Tile Publishing submodule is used to display heatmap overlays at a smooth frame rate on the construction site, and supports viewing on mobile or tablet devices. This submodule segments the heatmap into tiles, with tile sizes denoted as... (Unit: pixels), the number of levels in a multi-level pyramid is denoted as... (Dimensionless), and outputs PNG tiles for each layer; dynamically loaded on the client side according to the view zoom level to ensure smooth browsing even with low bandwidth. To avoid over-sharpening, the image is processed before output. Perform mild edge-preserving smoothing, with a window radius denoted as... (Unit: pixels).

[0219] Quality Assessment and Alerts: The Quality Assessment and Alerts submodule is used to check key quality quantities in real time during the generation phase: void percentage threshold. (Dimensionless), lower limit of coverage (Dimensionless) Upper limit of grid fitting condition number Consistency with global statistics. When an out-of-bounds error is detected, a "Data Sufficiency Warning" is displayed prominently on the report's homepage, and the out-of-bounds entry is added to the audit log. If necessary, a "Re-acquisition Suggestion" is issued via the processor to the synchronization control unit and the event image acquisition unit, with the suggested area marked as a semi-transparent polygon on the overlay layer.

[0220] Coordination with other units in the system: The results generation unit is used to: share mesh fitting and primary / secondary indices with the flatness evaluation unit to construct statistical tables; share dense height and invalid masks with the depth reconstruction unit to generate metric heatmaps; share coordinate transformations with the indoor coordinate establishment unit to ensure accurate overlay positions; share opening angle, vertex spacing, and pixel-length scaling factor with the cone parameter solving unit and the scale self-calibration unit to label parameters and units; share runtime quality and retrieval suggestions with the processor; and share overlay layers, statistical data, and audit logs with the storage unit for traceability and archiving.

[0221] Through the above structure and process, the result generation unit, with the combination of "coordinated organization + consistent color coding + statistical classification + standardized export + audit traceability + on-site slice display", can transform complex geometric and statistical information into clear, verifiable and archiveable engineering output.

[0222] It should be noted that the embodiments of the present invention have better implementability and are not intended to limit the present invention in any way. Any person skilled in the art may use the above-disclosed technical content to change or modify it into equivalent effective embodiments. However, any modifications or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A computer vision-based system for detecting the flatness of decorated walls, characterized in that, include: The system includes an event image acquisition unit, a biconical projection unit, a band separation unit, a synchronization control unit, a conical parameter solving unit, a scale self-calibration unit, a depth reconstruction unit, a flatness evaluation unit, a result generation unit, a storage unit, and a processor; among these, The processor is electrically connected to the storage unit, and is also signal-connected to the event image acquisition unit, the biconical projection unit, the band separation unit, the synchronization control unit, the conical parameter solving unit, the scale self-calibration unit, the depth reconstruction unit, the flatness evaluation unit, and the result generation unit, respectively. The dual-cone projection unit is used to simultaneously project a first and a second non-coplanar conical surface onto the wall surface being measured within a single scanning cycle. The opening angle of the first conical surface is... The opening angle of the second cone is The vertices of the two cones are a fixed, known distance apart along the projection optical axis. The opening angle , A positive quantity expressed in radians or degrees. A positive quantity expressed in units of length; The band separation unit is used to place the first conical surface and the second conical surface in a first band and a second band that do not overlap, respectively, and to provide a corresponding bandpass channel on the imaging side; The event image acquisition unit is used to record the event stream and its timestamp sequence triggered by the first cone and the second cone within the single scan cycle. ; The synchronization control unit is used to perform hardware-triggered synchronization between the event image acquisition unit and the double-cone projection unit, and to align the time index of the projection instruction sequence with the event timestamp. The cone parameter solving unit is used to solve the analytical parameters of the first cone and the second cone in the camera coordinate system. The analytical parameters include at least their respective opening angle, vertex position and axial direction. The scale self-calibration unit is used to base only on known distances. Opening angle , And the geometric quantities of the intersection lines of the two conical surfaces and the wall surface on the image plane, to calculate the pixel-length scaling factor. And establish the absolute dimensions of the wall surface, the pixel-length scaling factor Used to convert pixel distances in an image into physical length; The depth reconstruction unit is used for each pixel Calculate the two sets of heights given by the first and second cone surfaces. and It outputs the fusion height and effective pixel mask; The flatness evaluation unit is used to perform planar fitting on the fusion height in the indoor coordinate system and output the flatness index of the evaluation grid. The result generation unit is used to output an overlay layer registered with the wall image and a data file containing location indexes and statistics.

2. The computer vision-based wall surface flatness detection system according to claim 1, characterized in that, The dual-cone projection unit includes an axonocone lens assembly, an acousto-optic deflector, and a coaxial image relay mirror group; the acousto-optic deflector operates at a frequency within the same period. and Simultaneously, two coaxial ring beams are generated, which, after passing through the axial conical lens assembly, respectively produce opening angles of [missing information]. , The first conical surface and the second conical surface; the mechanical spacing of the coaxial image relay mirror group is a fixed known value. The distance between the vertices of the two conical surfaces along the projection optical axis is used to constrain the distance between them. And satisfy: in: , Indicates the driving frequencies of the two beams; The parameter indicating the upper limit of the opening angle difference; This indicates the installation and adjustment error of the vertex spacing; This indicates the upper limit of the allowable spacing error.

3. The computer vision-based wall surface flatness detection system according to claim 1, characterized in that, The band separation unit works in conjunction with the synchronization control unit to enable the event image acquisition unit to timestamp the event within a single scan cycle. Establish the pixel-cone index according to the following rules: Let the dual-channel time coding periods be respectively... and ,satisfy: when When the event belongs to the first cone; when Events are assigned to the second cone surface when both conditions are met; when both conditions are met, the smaller margin is used for assignment; where: , Indicates the time coding period of the two channels; Indicates the gate half-width setting value; This represents the operation of the greatest common divisor; This indicates the modulo operation.

4. The computer vision-based wall surface flatness detection system according to claim 1, characterized in that, The cone parameter solving unit is used to calculate the opening angle, vertex position and axial direction parameters of the first cone and the second cone respectively within a single scanning cycle, based on the time index provided by the synchronization control unit and the spatial distribution of the first event set and the second event set divided by the band separation unit on the image plane.

5. The computer vision-based wall surface flatness detection system according to claim 1, characterized in that, The scale self-calibration unit is used to utilize only known distances. Opening angle Opening angle The pixel-length scaling factor is analytically obtained from the set of intersection points of the two conical surfaces with the straight line features of the same wall surface on the image plane. and the The pixel-to-length scaling factor is written into the scale parameter table of the depth reconstruction unit. Used to convert pixel distance into physical length.

6. The computer vision-based wall surface flatness detection system according to claim 1, characterized in that, The depth reconstruction unit is used for each pixel The height obtained from the first and second cone surfaces is calculated above. and And based on self-calibration parameters Define an adaptive consistency threshold: when Output blending height: And calculate the normalized cone cost. and To meet ;when or The pixel is marked as invalid at that time; where, Indicates the basic threshold setting value; This indicates the set value of the weighting coefficient; , Represents the pixel weights calculated based on event density and bandwidth communication-to-noise ratio, and satisfying... ; This indicates the cost threshold setting value; Indicates the degree of integration; This represents the variable that minimizes the objective function.

7. The computer vision-based wall surface flatness detection system according to claim 3, characterized in that, The synchronization control unit is used to embed independent time codes into the driving sequence of the biconical projection unit and apply phase constraints: Simultaneously satisfy: To ensure time untangling within the same scanning cycle; the event image acquisition unit only acquires images with a width of The corresponding channel event is received within the gate control window; among which, , This indicates the phase offset of the two-channel time encoding; This indicates the time resolution period of the event acquisition link.

8. The computer vision-based wall surface flatness detection system according to claim 6, characterized in that, The flatness evaluation unit is used to evaluate the side length A square evaluation grid is used to perform a weighted least-squares fit on the fusion height to obtain a plane. The weight is It outputs a weighted flatness index. ; in, This indicates the grid scale factor setting value; Represents an interval-limiting function; This represents the set of valid pixels within the evaluation grid. , , Represents the fitted plane coefficients; This represents the weighted flatness index of the evaluation grid.

9. The computer vision-based wall surface flatness detection system according to claim 8, characterized in that, The system includes an indoor coordinate establishment unit, which is used to simultaneously utilize the wall-floor intersection direction vector. with respect to the direction vector of gravity Solve for coordinate alignment parameters and optimize through constraints: make Minimum to determine the wall normal Orientation in the indoor coordinate system; where, Represents the wall normal vector; Represents the unit vector of the wall-ground intersection line; This represents the unit vector indicating the direction of gravity. This indicates the weighting coefficient setting value.

10. The computer vision-based wall surface flatness detection system according to claim 9, characterized in that, The result generation unit is used to record parameter groups for each evaluation grid in the output overlay layer and data file: in, express The mean within this evaluation grid; , These represent the values ​​within the evaluation grid. The minimum and maximum values; The pixel-to-length scaling factor obtained from scale self-calibration is used to convert pixel distance into physical length; This represents the duty cycle vector for dual-channel events. These represent the ratios of the number of events in each of the two channels to the total number of events in the evaluation grid; the remaining symbols are consistent with the definitions in the preceding claims.

Images