A laser three-dimensional imaging system and imaging method

Abstract

This invention relates to the fields of optical measurement and three-dimensional imaging technology, specifically to a laser three-dimensional imaging system and method. The laser three-dimensional imaging system uses a VCSEL laser emission array as its light source, and its multiple channels employ a retinal-like logarithmic polar coordinate distribution. In the control system: a pre-sensing unit controls the activation of each channel in the light source to acquire surface information of the target at different resolutions and further converts it into a global feature density heatmap; an intelligent trajectory generation unit, based on a built-in feature density-sampling rate mapping relationship, combines it with the received global feature density heatmap to generate a variable-resolution scanning trajectory; and a point cloud fusion and reconstruction unit receives several three-dimensional point cloud data acquired based on the variable-resolution scanning trajectory and reconstructs a three-dimensional image. This invention achieves dynamic adaptive non-uniform variable-resolution imaging based on the target surface feature distribution, significantly improving overall scanning efficiency and reducing data redundancy while ensuring high accuracy in complex areas.

Description

Technical Field

[0001] This invention relates to the fields of optical measurement and three-dimensional imaging technology, specifically to a laser three-dimensional imaging system and imaging method. Background Technology

[0002] Laser 3D imaging technology obtains dense 3D spatial coordinate information of an object's surface by emitting laser beams and receiving reflected signals from the target. It is an indispensable core sensing method in fields such as online inspection in intelligent manufacturing, environmental perception in autonomous driving, and digital measurement in aerospace. With the increasing precision of industrial manufacturing and the growing complexity of autonomous driving scenarios, modern applications have placed stringent demands on 3D imaging, including the need for both large field-of-view coverage and local ultra-high resolution, as well as real-time response and efficient processing of massive amounts of data. Existing laser 3D imaging systems are mainly divided into three categories: first, line structured light scanners and time-of-flight radars based on fixed grid uniform sampling, with a globally fixed scanning density; second, solid-state lidar based on MEMS micromirrors or VCSEL arrays, where the resolution is determined by the physical parameters of the device and cannot be adjusted during operation; and third, multi-sensor fusion solutions, which combine wide-angle low-resolution systems with narrow field-of-view high-precision systems, but this adds extra hardware costs and registration complexity.

[0003] The aforementioned existing technologies have significant and interconnected drawbacks. First, the uniform scanning mechanism introduces an inherent contradiction between accuracy and speed. Scanning a 100mm×100mm area with a resolution of 0.1mm requires collecting millions of points and takes approximately 45 minutes, while flat areas account for over 90%, and key features only account for 5%-10%, with the vast majority of time wasted in areas with no information gain. Second, the hardware resolution is fixed; the MEMS galvanometer step angle and VCSEL light-emitting unit spacing are locked during the packaging stage. When the target contains both macroscopic structures and microscopic defects, it is forced to operate at the highest global resolution, resulting in significant resource waste. Third, the traditional three-stage process of pre-planning-execution-offline processing is serially fragmented and cannot dynamically adjust the strategy based on real-time scanning results. If anomalies are detected, a rescan is required, leading to low overall efficiency. Fourth, a large field of view and high resolution are mutually exclusive. A 45° field of view system has a resolution of only about 0.5mm at a working distance of 1m, while the field of view of a 0.05mm accuracy system is typically less than 5°, and multi-system stitching introduces additional registration errors. Fifth, uniform scanning generates massive amounts of redundant point clouds. A typical engine cylinder head scan produces approximately 50 million points, 90% of which are located on featureless planes. Processing a single frame takes several minutes, resulting in high storage and computational costs, which severely restricts real-time detection. Therefore, how to overcome the rigid constraints of uniform resolution within a single laser 3D imaging architecture and achieve non-uniform variable resolution imaging that adaptively adjusts the sampling density according to the target surface feature distribution, while simultaneously achieving high fidelity in key areas, high global efficiency, and low data redundancy, has become a pressing technical challenge in this field. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the defects existing in the prior art, thereby providing a laser three-dimensional imaging system and imaging method, specifically a laser three-dimensional imaging technology with a retinal-like non-uniform variable resolution mechanism.

[0005] A laser three-dimensional imaging system includes: an optical system and a control system. The optical system includes: a light source, an optical lens group, a MEMS micro-scanning mirror, a target under test, and a CMOS image sensor connected in sequence by optical paths. The light source uses a multi-channel VCSEL laser emission array, and the multi-channels are distributed in retinal-like logarithmic polar coordinates. The control system includes: a pre-sensing unit, an intelligent trajectory generation unit, and a point cloud fusion and reconstruction unit; The pre-sensing unit is connected to the light source and the CMOS image sensor. By controlling the activation of each channel in the light source, it acquires surface information of the target under test at different resolutions and further converts it into a global feature density heat map. The intelligent trajectory generation unit is connected to the pre-sensing unit and the MEMS micro-scanning mirror respectively. Based on the built-in feature density-sampling rate mapping relationship, the intelligent trajectory generation unit combines with the received global feature density heatmap to generate a variable resolution scanning trajectory; and further controls the operation of the MEMS micro-scanning mirror based on the variable resolution scanning trajectory. The point cloud fusion and reconstruction unit is connected to a CMOS image sensor, receives several three-dimensional point cloud data acquired based on a variable resolution scanning trajectory, and stitches the several three-dimensional point cloud data together using the iterative nearest point method to obtain the reconstructed three-dimensional image.

[0006] Preferably, the VCSEL laser emitting array has the following characteristics: from the center outwards, it consists of 32 channels in the center, 64 channels in the middle, and 32 channels on the periphery; the peak power of a single channel is 5mW, and the wavelength is 850nm±10nm. The center has 32 channels with a channel spacing of 5μm, corresponding to an imaging resolution of 2μm. The middle section has 64 channels, with the channel spacing increasing exponentially from the inside out, from 10μm to 40μm. It has 32 peripheral channels with a channel spacing of 80μm, corresponding to a resolution of 50μm.

[0007] Preferably, the workflow of the pre-sensing unit is as follows: The first global surface information of the target under test is obtained in a coarse manner with low resolution. Local surface information of complex regions in the first global surface information is obtained with medium resolution, and all information is processed to obtain a global feature density heatmap. In the first global surface information acquisition, only the outer 32 channels of the VCSEL laser emission array are enabled to construct a low-resolution parameter environment; When acquiring local surface information, only the middle 64 channels and the outer 32 channels are enabled to construct a medium-resolution parameter environment.

[0008] The preferred feature density-sampling rate mapping relationship is as follows: When entropy > 3.0, the sampling point spacing is 2μm; When 1.5 < entropy ≤ 3.0, the sampling point spacing is 10 μm; When the entropy is ≤1.5, the sampling point spacing is 50μm.

[0009] Preferably, the iterative nearest point method includes: coarse registration and fine registration; Coarse registration: Extract multiple pairs of low-resolution feature points from low-resolution point cloud data of several 3D point cloud datasets; coarsely stitch the multiple pairs of low-resolution feature points together to obtain the initial transformation matrix; Fine registration: Extracting high-resolution feature points from high-resolution point cloud data of several 3D point cloud datasets to optimize the initial transformation matrix.

[0010] Preferably, after obtaining the variable resolution scanning trajectory within the intelligent trajectory generation unit, the discrete resolution boundary is first fitted using a B-spline curve, and then the scanning sequence is optimized using the Dijkstra algorithm to obtain the optimal variable resolution scanning trajectory.

[0011] A laser three-dimensional imaging method is implemented using a laser three-dimensional imaging system.

[0012] The technical solution of this invention has the following advantages: This invention overcomes the precision-speed contradiction in traditional uniform scanning mechanisms in a single laser 3D imaging system, and realizes non-uniform variable resolution imaging that dynamically adapts to the distribution of target surface features. While ensuring high precision in complex areas, it significantly improves overall scanning efficiency and reduces data redundancy. Attached Figure Description

[0013] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0014] Figure 1 This is a system block diagram of the present invention; Figure 2 This is a schematic diagram of the channel distribution of a VCSEL laser emission array. Detailed Implementation

[0015] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0017] Example 1 like Figure 1 As shown, a laser three-dimensional imaging system includes: an optical system and a control system; The optical system includes: a light source, an optical lens group, a MEMS micro-scanning mirror, the target under test, and a CMOS image sensor, which are connected in sequence via optical paths; The light source uses a multi-channel VCSEL laser emission array, and the multi-channels are distributed in retinal-like logarithmic polar coordinates. VCSEL laser emitting array, each channel can be independently switched on / off and its intensity adjusted, specifically: such as Figure 2 From the center outwards, the channels are arranged as follows: 32 channels in the center, 64 channels in the middle, and 32 channels on the outer perimeter; peak power per channel: 5mW; wavelength: 850nm±10nm; modulation bandwidth ≥100MHz; supports PWM intensity modulation. The center has 32 channels with a channel spacing of 5μm, corresponding to an imaging resolution of 2μm. The middle section has 64 channels, with the channel spacing increasing exponentially from the inside out, from 10μm to 40μm. It has 32 peripheral channels with a channel spacing of 80μm, corresponding to a resolution of 50μm.

[0018] The optical lens assembly consists of a collimation module and a focusing module connected in sequence. It is used to collimate and focus the laser beam emitted by the VCSEL laser emitting array to achieve optical magnification adjustment. It should be noted that in practical applications, it is recommended that the parameters of the optical lens assembly be adapted to the application scenario. It is not recommended to use the same set of optical lens assemblies for different scenarios.

[0019] MEMS micro-scanning mirrors receive the laser beam output from the optical lens group and perform two-dimensional deflection scanning, covering the entire field of view. This provides the hardware foundation for large field-of-view imaging.

[0020] A CMOS image sensor receives laser signals reflected from the surface of the target object.

[0021] Example 2 Based on Example 1, this embodiment further discloses the composition of the control system; The control system includes: a pre-sensing unit, an intelligent trajectory generation unit, and a point cloud fusion and reconstruction unit; The pre-sensing unit is connected to the light source and the CMOS image sensor. By controlling the activation of each channel in the light source, it acquires surface information of the target at different resolutions and further converts it into a global feature density heatmap. The workflow of the pre-sensing unit is as follows: The first global surface information of the target under test is coarsely acquired at low resolution; the scanning speed is 1000 lines / second and the single frame scanning time is <1 second; the first global surface information includes: the overall outline of the target under test, the rough position of the target under test, and the preliminary positioning of complex areas on the target under test. Local surface information of complex regions in the first global surface information is obtained with medium resolution. All information is then processed to calculate the feature entropy, gradient magnitude, and curvature change of each region, which are used to obtain the feature density heatmap, thus obtaining the global feature density heatmap. The resolution is 256*256. It should be noted that calculating the feature density heatmap based on several indicator data is a conventional technique in this field. No improvement has been made in this embodiment. Those skilled in the art can select different indicators to calculate the heatmap according to their needs.

[0022] In the first global surface information acquisition, only the outer 32 channels of the VCSEL laser emission array are enabled to construct a low-resolution parameter environment; When acquiring local surface information, only the middle 64 channels and the outer 32 channels are enabled to construct a medium-resolution parameter environment, thereby improving the resolution to 10μm.

[0023] It should be noted that this embodiment uses hardware improvements to achieve variable resolution. Compared with the traditional downsampling scheme used in software, the amount of data collected is reduced by 80%, and the accuracy of 3D imaging of complex areas is improved to meet the requirements of precision inspection, making it suitable for the inspection of industrial precision parts.

[0024] The intelligent trajectory generation unit is connected to the pre-sensing unit and the MEMS micro-scanning mirror. Based on the built-in feature density-sampling rate mapping relationship, the intelligent trajectory generation unit combines this with the received global feature density heatmap to generate a variable-resolution scanning trajectory. Furthermore, it controls the operation of the MEMS micro-scanning mirror based on the variable-resolution scanning trajectory. The feature density-sampling rate mapping relationship is as follows: When entropy > 3.0, the sampling point spacing is 2μm; When 1.5 < entropy ≤ 3.0, the sampling point spacing is 10 μm; When entropy ≤ 1.5, the sampling point spacing is 50 μm. It should be noted that entropy measures the dispersion of a feature at a given resolution. A larger entropy value indicates a more uniform distribution of feature points in the feature space; a smaller entropy value indicates that feature points are more concentrated in complex regions.

[0025] After obtaining the variable resolution scanning trajectory within the intelligent trajectory generation unit, the discrete resolution boundary is first fitted using B-spline curves to avoid abrupt resolution changes. Then, the Dijkstra algorithm is used to optimize the scanning sequence to obtain the optimal variable resolution scanning trajectory, thereby reducing mechanical motion time.

[0026] The point cloud fusion and reconstruction unit is connected to a CMOS image sensor, receives several three-dimensional point cloud data acquired based on a variable resolution scanning trajectory, and stitches the several three-dimensional point cloud data together using the iterative nearest point method to obtain the reconstructed three-dimensional image.

[0027] It should be noted that in practical applications, since the resolution of the point cloud data at each location in the acquired 3D point cloud data is known; and since conventional image stitching methods require unifying the resolution of images with different resolutions, they are not suitable for the variable resolution 3D point cloud data of this embodiment. Therefore, in the embodiment, when seamlessly fusing 3D point cloud data using the iterative nearest point method, the coarse registration is as follows: first, extract multiple pairs of low-resolution feature points from the low-resolution point cloud data of the 3D point cloud data; then, coarsely stitch the multiple pairs of low-resolution feature points together to obtain the initial transformation matrix. Fine registration is achieved by optimizing the initial transformation matrix using high-resolution feature points extracted from high-resolution point cloud data of 3D point cloud data. The optimized initial transformation matrix is ​​applied to the source point cloud in the 3D point cloud data, updating its position to bring it closer to the target point cloud in the 3D point cloud data, until all point cloud data are stitched together to form the reconstructed 3D image. Registration accuracy is <0.5 minimum point spacing.

[0028] This embodiment also includes optimization of the 3D image: Smoothing of resolution transition areas: When stitching multi-resolution data, Gaussian weighted average is used to merge overlapping areas, and the width of the transition band is set to 5-10 voxel spacing to eliminate resolution steps and balance smoothness with detail preservation.

[0029] To address the trade-off between detail preservation and noise suppression in 3D surface reconstruction, this embodiment employs a partitioned adaptive reconstruction strategy. In target regions with high-resolution data, a Poisson surface reconstruction algorithm is introduced. By implicitly fitting an indicator function and solving the Poisson equation, the normal information of the high-density point cloud is fully utilized to accurately reconstruct subtle geometric features and sharp boundaries in the reconstructed surface. In background regions with low resolution or low signal-to-noise ratios, a moving least squares method is used for surface fitting. This method forms a smooth approximation surface by performing weighted multinomial regression on the point set within the local neighborhood, effectively suppressing random noise and outlier interference in the original data. The two methods are smoothly connected through a spatial mask, ultimately achieving differentiated reconstruction that preserves detail in fine-grained areas while denoising coarse-grained areas. This significantly reduces computational overhead and enhances local detail representation while maintaining the integrity of the overall model.

[0030] Example 3 A laser three-dimensional imaging method is implemented using a laser three-dimensional imaging system, as described in Examples 1-2.

[0031] This embodiment further compares the technical effects of the system disclosed in Embodiments 1-2 with existing scanning systems, as shown in Table 1; Table 1. Comparison of the imaging system performance of the present invention and the core performance indicators of the traditional uniform scanning system.

[0032] In practical applications, although the scanning time in the pre-sensing stage only accounts for 5-10% of the total scanning time, it can avoid 90% of invalid scans, with a feature point detection accuracy of ≥98% and a false negative rate of <0.5%.

[0033] Compared to traditional scanning path planning schemes, the present invention improves trajectory smoothness by 40%, reduces mechanical motion impact by 60%, and shortens the overall scanning path length by 30%. Furthermore, due to the close cooperation between various modules, this embodiment achieves sub-pixel level multi-scale registration accuracy in practical applications, and improves the surface quality index of the reconstructed model by 25%.

[0034] Example 4 Based on Example 3, this example further illustrates the testing of industrial components: Application Background: Precision industrial parts contain hundreds of tiny feature holes with diameters of 0.3-0.5mm. The accuracy and roundness of these holes directly affect equipment performance and lifespan. Traditional inspection methods require high-resolution scanning of the entire workpiece, which is extremely time-consuming.

[0035] System Configuration: VCSEL laser emitting array; Optical lens group: focal length 50mm, magnification 2x. MEMS micro-scanning mirror, scanning range 15 degrees × 15 degrees; Pre-sensing unit: FPGA; Intelligent trajectory planning unit: uses an ARM Cortex-A72 processor; Non-uniform fusion reconstruction unit: GPU image processor; CMOS image sensor, 5 megapixels, global shutter; The control system has its own signal processing circuit, which uses a 16-bit ADC for sampling at a sampling rate of 100MHz; The turbine under test is placed in the center of the scanning platform as the target to be inspected. Implementation steps: Step 1: Activate the pre-sensing unit: A global low-resolution rapid scan with a point spacing of 50μm is performed using a MEMS scanning mirror to obtain the overall outline of the target under test, perform preliminary feature detection, and locate the approximate position of all air film pores. Scanning of the air film pores and their edge regions: Further activation of the middle 64 channels to acquire local feature density; The feature indicators of each sub-region are calculated comprehensively to generate a feature density heatmap; Once pre-sensing is complete, a global feature density heatmap is output to the intelligent trajectory planning unit. Step 2: Resolution Decision Stage Air film pore region: entropy > 3.0, all 128 channels activated, highest resolution 2μm; Edge region of the target to be detected: 1.5 < feature entropy ≤ 3.0, 32 channels in the activation center + 64 channels in the middle, a total of 96 channels, resolution 10μm; Flat region of the target body to be detected: entropy ≤ 1.5; only the outer 32 channels are activated, resolution 50μm; Step 3: Intelligent Scanning Stage The intelligent trajectory planning unit generates an initial scanning path based on the resolution distribution; B-spline smoothing optimization is performed on the original discrete resolution boundary; A gradual sampling rate is used in the resolution transition band to avoid the step effect; Scan from high to low priority: air film pores - transition zone - target body to be detected; The scanning trajectory is optimized using a serpentine pattern to reduce mechanical idle travel; In this implementation scenario: a closed-loop feedback control unit is also provided to monitor the scanning quality in real time and automatically trigger a local rescan when the hole edge is found to be unclear; Step 4: Integration and Reconstruction Phase The non-uniform fusion reconstruction unit receives several point cloud data at different resolutions; Perform multi-scale iterative nearest-point registration: first, use low-resolution point cloud pairs for coarse registration, and then use high-resolution feature points for fine registration; Gaussian weighted average fusion is used in resolution transition regions to eliminate resolution steps; Poisson reconstruction is used in high-resolution regions to preserve fine features of the air film pores; Low-resolution regions are smoothed using moving least squares. Measure the diameter, roundness, and positional deviation of each air film pore and output an inspection report.

[0036] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A laser three-dimensional imaging system, comprising: The optical system and control system are characterized in that the optical system comprises: a light source, an optical lens group, a MEMS micro-scanning mirror, a target under test, and a CMOS image sensor connected in sequence by optical paths; The light source uses a multi-channel VCSEL laser emission array, and the multi-channels are distributed in retinal-like logarithmic polar coordinates. The control system includes: a pre-sensing unit, an intelligent trajectory generation unit, and a point cloud fusion and reconstruction unit; The pre-sensing unit is connected to the light source and the CMOS image sensor. By controlling the activation of each channel in the light source, it acquires surface information of the target under test at different resolutions and further converts it into a global feature density heat map. The intelligent trajectory generation unit is connected to the pre-sensing unit and the MEMS micro-scanning mirror respectively. Based on the built-in feature density-sampling rate mapping relationship, the intelligent trajectory generation unit combines with the received global feature density heatmap to generate a variable resolution scanning trajectory; and further controls the operation of the MEMS micro-scanning mirror based on the variable resolution scanning trajectory. The point cloud fusion and reconstruction unit is connected to a CMOS image sensor, receives several three-dimensional point cloud data acquired based on a variable resolution scanning trajectory, and stitches the several three-dimensional point cloud data together using the iterative nearest point method to obtain the reconstructed three-dimensional image.

2. The laser three-dimensional imaging system according to claim 1, characterized in that, The VCSEL laser emitting array, specifically, consists of 32 channels in the center, 64 channels in the middle, and 32 channels on the periphery, arranged from the center outwards; single-channel peak power: 5mW, wavelength: 850nm±10nm; The center has 32 channels with a channel spacing of 5μm, corresponding to an imaging resolution of 2μm. The middle section has 64 channels, with the channel spacing increasing exponentially from the inside out, from 10μm to 40μm. It has 32 peripheral channels with a channel spacing of 80μm, corresponding to a resolution of 50μm.

3. The laser three-dimensional imaging system according to claim 2, characterized in that, The workflow of the pre-sensing unit is as follows: The first global surface information of the target under test is obtained in a coarse manner with low resolution. Local surface information of complex regions in the first global surface information is obtained with medium resolution, and all information is processed to obtain a global feature density heatmap. In the first global surface information acquisition, only the outer 32 channels of the VCSEL laser emission array are enabled to construct a low-resolution parameter environment; When acquiring local surface information, only the middle 64 channels and the outer 32 channels are enabled to construct a medium-resolution parameter environment.

4. The laser three-dimensional imaging system according to claim 1, characterized in that, Feature density-sampling rate mapping relationship: When entropy > 3.0, the sampling point spacing is 2μm; When 1.5 < entropy ≤ 3.0, the sampling point spacing is 10 μm; When the entropy is ≤1.5, the sampling point spacing is 50μm.

5. A laser three-dimensional imaging system according to claim 1, characterized in that, Iterative closest point methods include: coarse registration and fine registration; Coarse registration: Extract multiple pairs of low-resolution feature points from low-resolution point cloud data of several 3D point cloud datasets; coarsely stitch the multiple pairs of low-resolution feature points together to obtain the initial transformation matrix; Fine registration: Extracting high-resolution feature points from high-resolution point cloud data of several 3D point cloud datasets to optimize the initial transformation matrix.

6. A laser three-dimensional imaging system according to claim 1, characterized in that, After obtaining the variable resolution scanning trajectory within the intelligent trajectory generation unit, the discrete resolution boundary is first fitted using B-spline curves, and then the scanning sequence is optimized using Dijkstra's algorithm to obtain the optimal variable resolution scanning trajectory.

7. A laser three-dimensional imaging method, characterized in that, It is implemented using a laser three-dimensional imaging system according to any one of claims 1-6.

Images