An eye feature acquisition and drug use identification device and method based on double optical path coaxial imaging

Abstract

The application provides an eye feature acquisition and drug use identification device and method based on double optical path coaxial imaging, which is provided with: a core imaging module, which adopts a double optical path structure sharing an entrance pupil to respectively acquire visible light and near-infrared eye images, ensuring that the two images are spatially parallax-free; a light supplementing module located on the object side of the core imaging module, which can selectively provide visible light and / or near-infrared illumination; a main control circuit module connected with the core imaging module and the light supplementing module, which is used for coordinating and controlling the operation timing and working state of each module, and executing a preset multi-modal acquisition logic; and a drug use judgment module built-in the main control circuit or an edge computing unit, which performs feature fusion analysis on multi-modal data and outputs a drug use judgment result. The application solves at least one of the technical problems in the prior art, i.e., the multi-modal data cannot be accurately aligned due to the parallax of double lenses, the light stimulation and imaging light path interfere with each other, and the accuracy of the traditional judgment algorithm is low.

Description

Technical Field

[0001] This invention relates to the technical field of physical detection and biometric identification, and in particular to a device and method for eye feature acquisition and drug use identification based on dual-optical-path coaxial imaging. Background Technology

[0002] Rapid screening of drug users is a crucial aspect of modern public safety governance and anti-drug efforts. Physical detection technologies based on eye characteristics offer advantages such as safety, speed, and reliability, and have replaced traditional biochemical testing methods such as urine, blood, hair, and saliva tests, becoming an important technical means for conducting rapid on-site screening.

[0003] However, in practical law enforcement operations and technological applications, existing multimodal eye detection devices still face technical bottlenecks in areas such as hardware imaging architecture and multi-source information fusion. Current technologies typically employ a discrete dual-lens architecture to acquire visible light and near-infrared images separately. However, this approach has drawbacks: due to the physical baseline distance between the visible light and infrared lenses, parallax inevitably occurs when photographing the human eye at close range. This parallax prevents the two images from precisely overlapping in spatial geometry, making pixel-level alignment of the pupil edge and iris texture difficult, thus limiting the accuracy of subsequent feature fusion based on deep learning. Furthermore, traditional devices often cannot address the problem of corneal reflections obscuring the pupil during visible light stimulation, and fixed-wavelength light sources are difficult to adapt to the specific needs of different ethnicities or covert law enforcement. Therefore, there is an urgent need to develop a dual-optical-path coaxial imaging device and method that can eliminate parallax at the physical optical path level, possesses "stimulus-imaging" anti-interference capabilities, and supports multi-dimensional feature fusion and analysis.

[0004] To address the aforementioned practical needs, this invention proposes an eye feature acquisition and drug use identification device and method based on dual-optical-path coaxial imaging. Through a compact optical architecture, it achieves parallax-free multimodal eye imaging, thereby effectively improving the accuracy and practical efficiency of rapid screening of drug users. Summary of the Invention

[0005] The purpose of this invention is to provide an eye feature acquisition and drug use identification device and method based on dual-optical-path coaxial imaging, which solves at least one of the technical problems in existing eye detection devices, namely, the inability to accurately align multimodal data due to dual-lens parallax, mutual interference between light stimulation and imaging optical paths, and low accuracy of traditional judgment algorithms.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: In a first aspect, the present invention provides an eye feature acquisition and drug use identification device based on dual-optical-path coaxial imaging, the device comprising: The core imaging module is configured with a dual-optical-path structure with a shared entrance pupil, used to acquire visible light and near-infrared images of the tested eye respectively, and the two images are spatially parallax-free. A supplementary lighting module is located on the object side of the core imaging module and is configured to selectively provide light beams in the visible light band and / or near-infrared band. The main control circuit module is connected to the core imaging module and the supplementary lighting module respectively, and is used to coordinate and control the running sequence and working state of each module, and execute the preset multimodal image data acquisition logic; The drug use assessment module is connected to the main control circuit module and is built into the main control circuit module or edge computing unit. It is configured to perform feature fusion and analysis based on the collected multimodal image data and output the drug use assessment result.

[0007] Furthermore, the core imaging module includes: The main lens module serves as a common light entry channel; A dichroic beam splitter is disposed on the image side of the main lens module and configured to separate incident light into a visible light beam that propagates along a first optical path and a near-infrared light beam that propagates along a second optical path. A first image sensor is disposed on the first optical path for acquiring images in the visible light band; The second image sensor, employing a global shutter design, is positioned on the second optical path to acquire the near-infrared band image.

[0008] Furthermore, the main lens module is equipped with an autofocus mechanism to drive the internal lens group to move along the optical axis to achieve autofocus.

[0009] Furthermore, the first image sensor has an infrared cutoff filter in front of its photosensitive area; the second image sensor has a narrowband filter in front of its photosensitive area.

[0010] Furthermore, the supplementary lighting module includes: A ring-shaped light source substrate on which several visible light emitting diodes and near-infrared light emitting diodes are arranged alternately to form an alternating light source array; A flexible light shield, shaped to fit the contour of the human eye socket, is integrated with the ring light source substrate to create a darkroom environment for the eye being tested during detection.

[0011] Furthermore, the main control circuit module is configured to execute the following stages of multimodal image data acquisition logic: Guidance and alignment stage: The near-infrared light source in the supplementary light module is controlled to be lit in the first power mode, and the core imaging module is driven to acquire preview images to guide the tested eye to complete the position alignment; Visible light static acquisition stage: Under the condition that the alignment is completed and no light stimulation is triggered, the visible light source in the supplementary light module is controlled to be lit, and the core imaging module is driven to acquire at least one frame of visible light static eye image; Near-infrared static reference acquisition stage: Under the condition of no visible light stimulation, the near-infrared light source in the supplementary light module is controlled to be lit in the second power mode, and the core imaging module is driven to acquire at least one frame of near-infrared static eye image. Dynamic response acquisition phase: While controlling the supplementary lighting module to provide a beam of light in the near-infrared band, the visible light source is controlled to be lit according to a preset stimulation sequence or intensity change pattern to provide light stimulation, and the core imaging module is synchronously driven to continuously acquire near-infrared eye image sequences.

[0012] Furthermore, at least one frame of visible light static eye image is acquired to obtain eye appearance features and initial pupil state information; At least one frame of near-infrared static eye image is acquired to obtain iris texture for identification, and pupil static state parameters before light stimulation are acquired as reference information for subsequent pupil dynamic analysis. Continuous acquisition of near-infrared eye image sequences is used to obtain dynamic changes in pupil state under light stimulation conditions.

[0013] Furthermore, the drug use assessment module includes: The input layer is configured to receive the multimodal image data; The feature extraction layer is configured to perform feature extraction and temporal parameter calculation on the multimodal image data, including: extracting multi-scale spatial texture features from the visible light static eye image and the near-infrared static eye image using a convolutional neural network; constructing a pupil diameter change curve over time based on the dynamic change sequence of the pupil state, and calculating quantitative feature parameters of the pupil physiological response based on the curve; The feature fusion layer is configured to fuse the texture features of the multi-scale space with the quantized features of the pupil physiological response to form a joint feature representation. The classification output layer is configured to output a classification result of whether the tested person has drug use behavior, or the corresponding drug use risk probability value, based on the joint feature representation.

[0014] Furthermore, the device also includes a human-computer interaction module; the human-computer interaction module includes: The display screen is used to show the picture-in-picture preview interface and detection results in real time; Physical button group, including buttons or knobs for triggering autofocus; The voice prompt unit is used to broadcast operation guidance instructions and test result warnings.

[0015] Secondly, the present invention also provides a method for eye feature acquisition and drug use identification based on dual-optical-path coaxial imaging, applied to the device described in any of the above claims, comprising the following steps: S1. Guidance and Alignment: Control the supplementary light module to provide near-infrared band illumination and drive the core imaging module to acquire preview images to guide the tested eye to complete position alignment; S2. Visible light static acquisition: Under the condition that the alignment is completed and no light stimulation is triggered, the supplementary light module is controlled to provide the visible light band, and the core imaging module is driven to acquire at least one frame of visible light static eye image; S3. Near-infrared static reference acquisition: Under the condition of no visible light stimulation, the supplementary light module is controlled to provide the near-infrared band, and the core imaging module is driven to acquire at least one frame of near-infrared static eye image; S4. Dynamic response acquisition: While controlling the supplementary light module to provide near-infrared band, it is also controlled to provide visible light stimulation according to the preset stimulation time sequence or intensity change mode, and the core imaging module is driven to continuously acquire near-infrared eye image sequences. S5. Multimodal feature extraction: Based on the visible light static eye image and the near-infrared static eye image, extract multi-scale spatial texture features; and based on the near-infrared eye image sequence, calculate the quantitative feature parameters of pupil physiological response; S6. Feature Fusion and Judgment: The multi-scale spatial texture features are fused with the quantitative feature parameters of the pupil physiological response, and based on the fused joint feature representation, the judgment result of whether the tested person has drug use behavior or the corresponding drug use risk probability value is output.

[0016] Compared with the prior art, the present invention has at least the following beneficial effects: (1) Parallax-free precise fusion: At the hardware level, the core imaging module constructs a dual-optical-path structure with a shared entrance pupil through the main lens module and the dichroic beam splitter, so that the visible light and near-infrared images are completely parallax-free in space. This eliminates the geometric misalignment defect caused by the baseline distance of traditional dual lenses from the root of the physical optical path, and provides the original data basis for pixel-level precise registration for subsequent multimodal feature fusion, effectively improving the accuracy of feature extraction and matching.

[0017] (2) Improve measurement accuracy and anti-interference capability: The main control circuit module executes the phased control logic of “guidance and alignment - visible light static acquisition - near-infrared static reference acquisition - dynamic response acquisition”. In the dynamic response acquisition stage, the near-infrared band beam is used as the imaging light and the visible light band beam is used as the stimulation light. The narrow band filter in front of the near-infrared sensor effectively suppresses the reflection of the visible light band, realizing the spectral separation of the stimulation light and the imaging light. This results in a pure pupil contraction process image with complete edges and no reflective light spots, which significantly improves the accuracy and anti-interference capability of pupil dynamic measurement.

[0018] (3) Intelligent judgment is efficient: At the level of intelligent judgment, a convolutional neural network algorithm / hierarchical feature modeling architecture is introduced to deeply integrate and analyze static texture (identity) features with dynamic pupil physiological response features, which makes up for the lack of information dimension of the traditional single threshold judgment method, significantly reduces the false alarm rate, and realizes the dual accurate locking of the identity and physiological state of drug users. Attached Figure Description

[0019] 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.

[0020] Figure 1 This is a schematic diagram of the optical path principle of the dual-optical-path coaxial optical module provided in this embodiment; Figure 2 This is a side view of the internal structure of the detection device provided in this embodiment; Figure 3 This is a front view of the detection device provided in this embodiment; Figure 4 This is a rear view of the detection device provided in this embodiment; Figure 5 This is a top view of the detection equipment provided in this embodiment; Figure 6 This is a schematic diagram of the image acquisition process of the detection device provided in this embodiment.

[0021] Figure label: 1-Main lens; 2-Dichroic beam splitter; 3-Infrared cut-off filter; 4-First image sensor; 5-Narrow band filter; 6-Second image sensor; 7-Removable clip; 8-Lens hood; 9-Ring light source substrate; 10-Power button; 11-Function button; 12-Auto focus button; 13-Display screen. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. 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.

[0023] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0024] I. Examples This embodiment provides an eye feature acquisition and drug use identification device based on dual-optical-path coaxial imaging. Please refer to... Figure 1-3 As shown, the system includes a core imaging module, a supplementary lighting module, a main control circuit module, and a drug use assessment module. The core imaging module employs a dual-path structure with a shared entrance pupil, acquiring visible light and near-infrared images separately, ensuring complete parallax elimination in spatial geometry between the two images, thus laying the hardware foundation for subsequent multimodal data fusion. The supplementary lighting module, located on the object side of the core imaging module, can selectively provide visible light and / or near-infrared beams as needed to meet the supplementary lighting requirements at different acquisition stages. The main control circuit module is electrically connected to each module, responsible for coordinating and controlling the operational sequence of imaging and supplementary lighting, as well as executing preset multimodal image acquisition logic. The drug use assessment module is built into the main control circuit or edge computing unit, performing feature extraction, fusion analysis, and ultimately outputting the drug use assessment result based on the acquired visible light and near-infrared static and dynamic sequence images. This architecture eliminates parallax interference at the physical optics level and achieves precise coordination between light stimulation and imaging through modular design, significantly improving the accuracy of drug screening in complex scenarios and the equipment's environmental adaptability.

[0025] This embodiment defines the specific structure of the core imaging module, including a main lens module, a dichroic beam splitter, a first image sensor 4, and a second image sensor 6. The main lens module serves as a common entrance light channel, ensuring that visible light and near-infrared light enter the optical system along the same optical axis. The dichroic beam splitter is located on the image side of the main lens 1, and can specifically be a dichroic beam splitter prism 2 or a similar product, separating the incident light in the spectral dimension. For example, it transmits a visible light beam to the first optical path and reflects a near-infrared beam to the second optical path. The first image sensor 4 is located on the first optical path and is used to acquire visible light images; the second image sensor 6 is located on the second optical path and employs a global shutter design to eliminate pupil motion blur, and is used to acquire near-infrared images. Since the two sensors share the same entrance pupil, the two images are spatially parallax-free, solving the image geometric misalignment problem caused by baseline distance in traditional discrete dual-lens solutions, and ensuring spatial consistency during subsequent multimodal feature fusion.

[0026] This embodiment further specifies that the main lens module is equipped with an autofocus mechanism, capable of driving the internal lens group to move along the optical axis to achieve autofocus. In specific implementation, when the human-machine interface module issues a focus command or the system automatically determines that focus is needed, the main control circuit module analyzes the preview image in real time according to the image sharpness evaluation function, and controls actuators such as the voice coil motor to drive the lens group to fine-tune its position until the sharpness function reaches its peak. This design ensures that the core imaging module can obtain clear iris texture images and pupil edge contours under different shooting distances and different deviations in the eye position of the subject, providing reliable raw image data for subsequent high-precision pupil diameter measurement and texture feature extraction, and avoiding feature blurring or measurement errors caused by defocus.

[0027] This embodiment defines the filter configurations for the two image sensors. The first image sensor 4 (visible light channel) has an infrared cutoff filter 3 in front of its photosensitive area. This filter removes stray light in the near-infrared band, preventing infrared noise from interfering with the color and detail reproduction of the visible light image and ensuring clear representation of the visible light image's appearance features. The second image sensor 6 (near-infrared channel) has a narrow-band filter 5 in front of its photosensitive area. For example, a bandpass filter with a center wavelength of 850nm and a narrow half-width allows only specified near-infrared light to enter the sensor, significantly suppressing corneal reflections and other environmental stray light generated during visible light stimulation. This design is a key hardware support for achieving the "visible light stimulation, near-infrared imaging" separation mode. It effectively solves the problem of strong visible light stimulation causing reflections on the corneal surface that obstruct the pupil, ensuring that the near-infrared sensor captures images of the pupil contraction process with complete edges and no light spot obstruction, thereby improving the accuracy of dynamic pupil measurement.

[0028] This embodiment refines the structure of the supplementary lighting module, including a ring-shaped light source substrate 9 and a flexible light shield 8. The ring-shaped light source substrate 9 has several visible light-emitting diodes (LEDs) and near-infrared LEDs arranged alternately to form an alternating light source array. This layout ensures that the optical axes of visible light and near-infrared illumination are aligned at the same height, avoiding uneven illumination caused by unilateral shadows. Furthermore, the ring structure adapts to the shape around the eye, providing a uniform surface light source. The flexible light shield 8 is shaped to fit the contour of the human eye socket and is integrated with the ring-shaped light source substrate 9. During use, it conforms to the eye socket area of ​​the test subject, effectively isolating external ambient light and creating a relatively stable dark chamber environment for the human eye during testing. This dark chamber condition ensures that the pupil's response to light stimulation is not interfered with by stray light, resulting in a stable initial dark adaptation state of the pupil and comparable dynamic response curves. Simultaneously, the flexible material improves the wearing comfort of the test subject.

[0029] This embodiment details the multimodal image data acquisition logic executed by the main control circuit module, which is divided into the following stages: Guidance and alignment stage: The near-infrared light source is controlled to illuminate in a low-power (relative to the power consumption of the second power / normal operating power) first power mode and a preview image is acquired. This provides invisible infrared illumination for alignment preview while avoiding premature stimulation of pupil contraction. Visible light static acquisition stage: The visible light source is illuminated under no light stimulation conditions, and at least one frame of visible light static eye image is acquired, recording the initial state of the pupil and the appearance features of the eye. Near-infrared static reference acquisition stage: Under no visible light stimulation conditions, the near-infrared light source is switched to a higher power (relative to the aforementioned low power) second power mode, and at least one frame of near-infrared static eye image is acquired, obtaining high-resolution iris texture and the static reference parameters of the pupil before stimulation. Dynamic response acquisition stage: Under continuous near-infrared illumination conditions, the visible light source is illuminated according to a preset visible light stimulation timing or intensity change pattern, and a near-infrared eye image sequence is acquired synchronously and continuously, recording the dynamic changes of the pupil under light stimulation. This phased timing control logic scientifically distinguishes between static baseline acquisition and dynamic response acquisition, ensuring the accuracy of pre-stimulation baseline data, and achieving interference-free dynamic response monitoring through spectral separation of stimulus light and imaging light.

[0030] This embodiment clarifies the specific purpose and data significance of the images acquired at each stage. At least one frame of visible light static eye image is acquired to obtain appearance features and initial pupil state information before stimulation; these characteristics can serve as auxiliary evidence for drug use assessment. At least one frame of near-infrared static eye image is acquired to extract iris texture details for identification and to measure pupil static state parameters before stimulation (such as the baseline pupil diameter under dark adaptation), serving as an important reference for subsequent dynamic analysis. The continuously acquired near-infrared eye image sequence completely records the pupil's changes under visible light stimulation, providing a data source for subsequent construction of pupil diameter-time curves and calculation of physiological response quantification parameters. This design ensures that each image frame has a clear functional positioning, and the synergistic effect of multi-dimensional data enhances the specificity and sensitivity of drug use detection.

[0031] This embodiment describes the internal hierarchical architecture of the drug use assessment module, including an input layer, a feature extraction layer, a feature fusion layer, and a classification output layer. The input layer receives visible light static eye images, near-infrared static eye images, and near-infrared eye image sequences from the aforementioned acquisition phase. The feature extraction layer uses a convolutional neural network to extract multi-scale spatial texture features from the static images, capturing fine iris texture and abnormal eye structures; simultaneously, it constructs a pupil diameter change curve over time based on the dynamic image sequence and automatically calculates a series of quantitative pupil physiological response feature parameters from this curve. The feature fusion layer concatenates, weights, or maps the texture feature vector output by the convolutional neural network with the quantitative pupil physiological response feature vector to form a unified joint feature representation, which simultaneously encodes biological anatomical features and neurophysiological response features. The classification output layer, based on the joint feature representation, outputs the classification result of drug use behavior or the probability value of drug use risk through a fully connected network or classifier. This hierarchical architecture fully utilizes the complementarity of multimodal information, significantly reducing the false alarm rate compared to single feature threshold judgment methods, and improving the intelligence and reliability of drug use screening.

[0032] This embodiment further defines the device as including a human-machine interface module, which integrates a display screen 13, a physical button group, and a voice prompt unit. The display screen 13 presents a picture-in-picture preview interface in real time, simultaneously displaying a visible light positioning map and a near-infrared texture map, helping the operator intuitively confirm the alignment and instantly display the detection results. The physical button group includes buttons or knobs for triggering autofocus, supporting focusing, acquisition, and other operation modes, facilitating quick operation by law enforcement personnel in handheld scenarios. The voice prompt unit broadcasts operation guidance instructions and detection result warnings, reducing reliance on the operator's professional skills and improving the convenience and user-friendliness of on-site law enforcement. This module design fully considers the application scenario of rapid on-site screening, effectively improving the ergonomics and practicality of the equipment.

[0033] This embodiment also provides a method for eye feature acquisition and drug use identification using dual-path coaxial imaging applied to the aforementioned device, comprising the following steps: Step 1: Guidance and Alignment. The supplementary lighting module provides near-infrared illumination and acquires preview images to guide the eye under test to complete positional alignment. Step 2: Visible Light Static Acquisition. Visible light illumination is provided under no-light-stimulation conditions, and a static visible light eye image is acquired. Step 3: Near-Infrared Static Reference Acquisition. Near-infrared illumination is provided under no-visible-light-stimulation conditions, and a static near-infrared eye image is acquired. Step 4: Dynamic Response Acquisition. While maintaining near-infrared illumination, visible light stimulation is provided according to a preset stimulation mode, and a sequence of near-infrared eye images is acquired synchronously and continuously. Step 5: Multimodal Feature Extraction. Multi-scale spatial texture features are extracted from the static images, and quantitative feature parameters of pupil physiological response are calculated from the dynamic sequence. Step 6: Feature Fusion and Judgment. Texture features and physiological features are fused to output a drug use behavior judgment result or risk probability value. This method automates the entire process from alignment, benchmark acquisition, stimulus response recording to intelligent analysis through a strictly controlled time-series process and spectral separation strategy. It effectively overcomes the shortcomings of traditional methods such as parallax interference, reflective occlusion, and single feature, providing a high-precision and high-reliability technical means for rapid initial screening of drug users.

[0034] II. Specific Implementation (I) Hardware Architecture of Dual-Path Coaxial Imaging Equipment like Figure 1 (Optical path schematic diagram) Figure 2 (Internal structure side view) and Figure 3 As shown in the front view, this embodiment provides an eye feature acquisition and drug identification device based on dual-optical-path coaxial imaging. Its main body is a handheld structure, which mainly includes a core imaging module, a replaceable supplementary light module, a human-computer interaction module and a main control circuit module.

[0035] 1. Core imaging module (coaxial optical path system) such as Figure 1 and Figure 2 As shown, this module is key to achieving parallax-free imaging, and is distributed sequentially from the object side to the image side along the optical axis: Main lens module: Serving as a common light-entry channel, it integrates the main lens 1 and an autofocus mechanism driven by a voice coil motor. This mechanism connects to the main control circuit module and can drive the lens group to fine-tune along the optical axis according to the image sharpness evaluation function, ensuring clear images at different shooting distances.

[0036] Dichroic beam splitter 2: Located behind the main lens 1. Its beam splitting surface is coated with a special dielectric film system, configured to transmit visible light of 400nm-700nm and reflect near-infrared light of 700nm-1100nm (or vice versa), thereby separating the incident light into two independent paths without loss in the spectral dimension.

[0037] First image sensor 4 (visible light channel): It is set in the transmission light path of the prism, and an infrared cut-off filter 3 is attached in front of its photosensitive front to filter out infrared noise and generate a high-fidelity visible light image of the eye appearance (such as for auxiliary positioning or scleral feature analysis).

[0038] The second image sensor 6 (near-infrared channel) is located on the reflected light path of the prism, and a narrow-band filter 5 with a center wavelength of 850nm (half-wavelength ±10nm) is provided in front of its photosensitive front.

[0039] Technical Feature Optimization: To meet the requirements of capturing rapid pupil contraction in drug detection, this sensor specifically uses a global shutter CMOS sensor. Compared to traditional rolling shutters, the global shutter ensures that all pixels in a single frame are exposed simultaneously, completely eliminating the "jelly effect" and motion blur caused by handheld shooting or high-speed pupil movement, ensuring the sharpness of iris texture and pupil edges.

[0040] Coaxial advantage: Since the two sensors share the same entrance pupil of the main lens 1, their observation angles of the tested eye are completely overlapped, which physically eliminates parallax and provides a hardware foundation for subsequent multimodal image pixel-level registration.

[0041] 2. Replaceable fill light modules, such as Figure 2 and Figure 3 As shown, this module is installed at the front of the device and adopts an integrated packaging structure.

[0042] Structural components: including a flexible silicone light shield 8 adapted to the contour of the human eye socket (for creating a darkroom environment) and an embedded ring light source substrate 9.

[0043] Alternating light source array: On a ring substrate, visible light LEDs and near-infrared LEDs are arranged alternately. This layout ensures that the illumination centers of the two spectra are aligned with the optical axis, avoiding unilateral shading.

[0044] Quick-release and wavelength switching: The bottom of the module features a quick-release connection mechanism (such as a detachable clip 7 or a magnetic structure) and gold-plated electrical contacts. Users can replace the module according to their actual needs. Scenario A (Standard): Use the standard 850nm infrared fill light module to obtain the clearest iris texture; Scenario B (concealed / sensitive): Replace with a 940nm infrared supplementary light module (without red burst) to avoid the subject noticing or experiencing photo-stress response.

[0045] 3. Human-computer interaction module, such as Figure 4 (Rear view) and Figure 5 As shown in the top view, interactive components are located on the back and top of the device: Display (touch) screen: used to display in real time the picture-in-picture preview interface (visible light positioning map + infrared texture map) synthesized by the main control module and the final detection results.

[0046] Physical button group: including the autofocus button on the top (supports half-press for focusing and full-press for acquisition), the mode switching button on the side (quick screening / precise identification), and the power reset button.

[0047] Voice prompt unit: Built-in speaker for broadcasting voice commands such as "Please look at the camera", "Detection in progress", "Pass", etc., reducing the operating threshold.

[0048] (II) Control Logic and Data Acquisition Process like Figure 6 As shown, the main control circuit module in this embodiment is configured to execute a time-sensitive "photostimulation-response" acquisition process, the specific steps of which are as follows: S1. Guidance and Alignment (Preview Mode) After the main control module recognizes the user's operation, it controls the near-infrared LED in the supplementary lighting module to work in a low-power constant-on mode.

[0049] The second image sensor 6 (infrared) utilizes its high sensitivity to capture real-time images and displays a preview on the screen. At this time, the infrared light intensity is extremely low and invisible to the human eye, and will not cause pupil constriction. It is mainly used to assist the operator in adjusting the position of the device until the screen indicator is aligned with the eyes.

[0050] S2. Visible light static acquisition (visible light static eye image) After alignment, the main control module briefly illuminates the visible light LED to drive the first image sensor 4 (visible light) to acquire a single frame of static image. This image is primarily used to record the external features of the eye (ptosis, conjunctival congestion, and other signs of drug use) and to assist in confirming the initial positioning of the pupil.

[0051] S3. Near-infrared baseline acquisition (dark adaptation state, near-infrared static eye image) The main control module turns off the visible light LED and switches the near-infrared LED to high-power shooting mode.

[0052] The second image sensor 6 (infrared) is driven to acquire multiple clear static images of the iris and the pupil before stimulation.

[0053] Data significance: At this time, due to the obstruction of light shield 8 and the absence of visible light stimulation, the pupil is in a natural dark-adapted dilation state. The system calculates the pupil diameter at this time as a "baseline value," and simultaneously uses this image to extract high-precision iris texture for identity recognition.

[0054] S4. Dynamic response acquisition (stimulus-imaging separation, near-infrared eye image sequence) Light stimulation: The main control module controls the visible light LED to emit a strong light stimulation signal with a specific timing sequence.

[0055] Synchronous acquisition: At the same time as the visible light stimulus is emitted, the main control module synchronously triggers the second image sensor 6 (infrared) to acquire a continuous video stream at a high frame rate.

[0056] Key advantage: Utilizing infrared light to capture pupillary responses to visible light stimulation. Because the imaging light (infrared) and the stimulation light (visible) have different wavelengths, and there is a narrow-band filter 5 in front of the sensor, the strong reflective spot formed on the cornea by visible light stimulation will not be imaged by the infrared sensor, thus obtaining a pure image of the pupillary contraction process with complete edges and no light spot obstruction.

[0057] S5. Multimodal Feature Extraction Based on the visible light static eye image and the near-infrared static eye image, multi-scale spatial texture features are extracted; and based on the near-infrared eye image sequence, quantitative feature parameters of pupil physiological response are calculated.

[0058] S6. Feature Fusion and Analysis The multi-scale spatial texture features are fused with the quantitative feature parameters of the pupil physiological response, and based on the fused joint feature representation, the result of the judgment of whether the tested person has drug use behavior or the corresponding drug use risk probability value is output.

[0059] (III) Drug Use Analysis Module and Algorithm Architecture In this embodiment, the drug use analysis module runs on the device's built-in edge computing unit and uses a layered deep learning architecture to process the collected multimodal data. 1. Input Layer This layer is configured to receive full-dimensional, multi-modal raw data acquired based on the aforementioned control logic, specifically including three independent data streams: Data stream A (visible light static information): Receives a visible light static eye image from step S2. This image contains external characterization information such as ptosis, conjunctival hyperemia, and nystagmus, which is an important auxiliary basis for determining drug use.

[0060] Data stream B (Near-infrared static information): Receives the near-infrared static iris image from step S3. This image, captured in the 850nm / 940nm band, features high-contrast texture details and is used for identity feature extraction and measurement of dark-adapted pupil reference values.

[0061] Data stream C (Dynamic Physiological Information): Receives the sequence of dynamic changes in pupil state from step S4. This sequence records the complete spatiotemporal changes in the pupil from constriction to recovery under visible light stimulation.

[0062] 2. Feature Extraction Layer This layer is configured to perform deep analysis on the input multimodal raw data, specifically divided into two parallel processing branches: static texture and dynamic physiological data. Identity and Static Feature Branches: Input data: Receive the visible light static eye image acquired in step S2 and the near-infrared static iris image acquired in step S3.

[0063] Processing logic: A lightweight convolutional neural network is used to perform convolution operations on the above images. The network is configured to extract multi-scale spatial texture features.

[0064] Output features: On the one hand, high-dimensional iris texture feature vectors are extracted for accurate comparison of identity IDs; on the other hand, visible light image features are used to detect whether there are abnormal structural features in the eyes, forming static auxiliary criteria.

[0065] Dynamic physiological characteristics branch: Input data: The sequence of dynamic changes in pupil state acquired under visible light stimulation conditions in step S4.

[0066] Processing logic: Perform frame-level temporal analysis on the dynamic sequence of pupils to construct a high-precision curve of pupil diameter change over time.

[0067] Quantitative parameter calculation: The algorithm automatically calculates a set of key physiological response quantitative characteristic parameters based on the PLR ​​curve to comprehensively characterize the physical properties of the pupil's light response. These parameters include, but are not limited to: Response delay time: the time difference between the onset of light stimulation and the onset of pupil constriction; Maximum contraction rate: The maximum rate of change during pupil contraction; Maximum growth rate / recovery rate: The maximum rate at which the pupil dilates after the stimulation ends; Pupil constriction ratio: the ratio of the smallest diameter after constriction to the reference diameter before constriction; Recovery time: The time required for the pupil to return to a specific proportional diameter; Extreme parameters: morphological indicators such as maximum scale, minimum scale, and scale range.

[0068] 3. Feature Fusion and Classification Layer This layer is responsible for transforming heterogeneous feature data into the final judgment result. The specific processing logic is as follows: Feature fusion strategy: Employ splicing, weighting, or mapping fusion strategies.

[0069] Specific operation: Combine the high-dimensional spatial texture feature vector output by the identity and static feature branches with the pupil physiological response quantization feature vector output by the dynamic physiological feature branches.

[0070] Objective: To construct a unified joint feature representation. This feature representation simultaneously encodes the biological anatomical features and neurophysiological response features of the tested object, achieving deep complementarity between physical and physiological features.

[0071] Classification Output: The joint feature representation is input into a fully connected layer or support vector machine classifier, and the output consists of two dimensions of evaluation metrics: Classification results: Output a binary label indicating whether the tested person has or does not engage in drug use, or a specific drug type preference; Drug use risk probability value: Outputs a continuous confidence value from 0 to 100% to quantify the reliability of the current judgment and assist law enforcement officers in making decisions.

[0072] In summary, this embodiment, through the hardware design of dual optical paths coaxial and global shutter, combined with the software's "stimulus-imaging separation" control logic and CNN depth analysis, successfully solves the problems of large parallax, strong reflective interference, and low recognition rate of traditional equipment, and achieves integrated high-precision screening of drug users.

Claims

1. A device for eye feature acquisition and drug use identification based on dual-optical-path coaxial imaging, characterized in that, The device is equipped with: The core imaging module is configured with a dual-optical-path structure with a shared entrance pupil, used to acquire visible light and near-infrared images of the tested eye respectively, and the two images are spatially parallax-free. A supplementary lighting module is located on the object side of the core imaging module and is configured to selectively provide light beams in the visible light band and / or near-infrared band. The main control circuit module is connected to the core imaging module and the supplementary lighting module respectively, and is used to coordinate and control the running sequence and working state of each module, and execute the preset multimodal image data acquisition logic; The drug use assessment module is connected to the main control circuit module and is built into the main control circuit module or edge computing unit. It is configured to perform feature fusion and analysis based on the collected multimodal image data and output the drug use assessment result.

2. The eye feature acquisition and drug use identification device based on dual-optical-path coaxial imaging according to claim 1, characterized in that, The core imaging module includes: The main lens module serves as a common light entry channel; A dichroic beam splitter is disposed on the image side of the main lens module and configured to separate incident light into a visible light beam that propagates along a first optical path and a near-infrared light beam that propagates along a second optical path. A first image sensor is disposed on the first optical path for acquiring images in the visible light band; The second image sensor, employing a global shutter design, is positioned on the second optical path to acquire the near-infrared band image.

3. The eye feature acquisition and drug use identification device based on dual-optical-path coaxial imaging according to claim 2, characterized in that, The main lens module is equipped with an autofocus mechanism, which drives the internal lens group to move along the optical axis to achieve autofocus.

4. The eye feature acquisition and drug use identification device based on dual-optical-path coaxial imaging according to claim 2, characterized in that, The first image sensor has an infrared cutoff filter in front of its photosensitive area; the second image sensor has a narrowband filter in front of its photosensitive area.

5. The eye feature acquisition and drug use identification device based on dual-optical-path coaxial imaging according to claim 1, characterized in that, The supplementary lighting module includes: A ring-shaped light source substrate on which several visible light emitting diodes and near-infrared light emitting diodes are arranged alternately to form an alternating light source array; A flexible light shield, shaped to fit the contour of the human eye socket, is integrated with the ring light source substrate to create a darkroom environment for the eye being tested during detection.

6. The eye feature acquisition and drug use identification device based on dual-optical-path coaxial imaging according to claim 1, characterized in that, The main control circuit module is configured to execute the following stages of multimodal image data acquisition logic: Guidance and alignment stage: The near-infrared light source in the supplementary light module is controlled to be lit in the first power mode, and the core imaging module is driven to acquire preview images to guide the tested eye to complete the position alignment; Visible light static acquisition stage: Under the condition that the alignment is completed and no light stimulation is triggered, the visible light source in the supplementary light module is controlled to be lit, and the core imaging module is driven to acquire at least one frame of visible light static eye image; Near-infrared static reference acquisition stage: Under the condition of no visible light stimulation, the near-infrared light source in the supplementary light module is controlled to be lit in the second power mode, and the core imaging module is driven to acquire at least one frame of near-infrared static eye image. Dynamic response acquisition phase: While controlling the supplementary lighting module to provide a beam of light in the near-infrared band, the visible light source is controlled to be lit according to a preset stimulation sequence or intensity change pattern to provide light stimulation, and the core imaging module is synchronously driven to continuously acquire near-infrared eye image sequences.

7. The eye feature acquisition and drug use identification device based on dual-optical-path coaxial imaging according to claim 6, characterized in that, Acquire at least one frame of visible light static eye image to obtain eye appearance features and initial pupil state information; At least one frame of near-infrared static eye image is acquired to obtain iris texture for identification, and pupil static state parameters before light stimulation are acquired as reference information for subsequent pupil dynamic analysis. Continuous acquisition of near-infrared eye image sequences is used to obtain dynamic changes in pupil state under light stimulation conditions.

8. The eye feature acquisition and drug use identification device based on dual-optical-path coaxial imaging according to claim 6, characterized in that, The drug use assessment module includes: The input layer is configured to receive the multimodal image data; The feature extraction layer is configured to perform feature extraction and temporal parameter calculation on the multimodal image data, including: extracting multi-scale spatial texture features from the visible light static eye image and the near-infrared static eye image using a convolutional neural network; constructing a pupil diameter change curve over time based on the dynamic change sequence of the pupil state, and calculating quantitative feature parameters of the pupil physiological response based on the curve; The feature fusion layer is configured to fuse the texture features of the multi-scale space with the quantized features of the pupil physiological response to form a joint feature representation. The classification output layer is configured to output a classification result of whether the tested person has drug use behavior, or the corresponding drug use risk probability value, based on the joint feature representation.

9. The eye feature acquisition and drug use identification device based on dual-optical-path coaxial imaging according to claim 1, characterized in that, The device also includes a human-computer interaction module; the human-computer interaction module includes: The display screen is used to show the picture-in-picture preview interface and detection results in real time; Physical button group, including buttons or knobs for triggering autofocus; The voice prompt unit is used to broadcast operation guidance instructions and test result warnings.

10. A method for eye feature acquisition and drug use identification based on dual-optical-path coaxial imaging, applied to the device according to any one of claims 1 to 9, characterized in that, Includes the following steps: S1. Guidance and Alignment: Control the supplementary light module to provide near-infrared band illumination and drive the core imaging module to acquire preview images to guide the tested eye to complete position alignment; S2. Visible light static acquisition: Under the condition that the alignment is completed and no light stimulation is triggered, the supplementary light module is controlled to provide the visible light band, and the core imaging module is driven to acquire at least one frame of visible light static eye image; S3. Near-infrared static reference acquisition: Under the condition of no visible light stimulation, the supplementary light module is controlled to provide the near-infrared band, and the core imaging module is driven to acquire at least one frame of near-infrared static eye image; S4. Dynamic response acquisition: While controlling the supplementary light module to provide near-infrared band, it also controls it to provide visible light stimulation according to the preset stimulation time sequence or intensity change mode, and synchronously drives the core imaging module to continuously acquire near-infrared eye image sequences. S5. Multimodal feature extraction: Based on the visible light static eye image and the near-infrared static eye image, extract multi-scale spatial texture features; and based on the near-infrared eye image sequence, calculate the quantitative feature parameters of pupil physiological response; S6. Feature Fusion and Judgment: The multi-scale spatial texture features are fused with the quantitative feature parameters of the pupil physiological response, and based on the fused joint feature representation, the judgment result of whether the tested person has drug use behavior or the corresponding drug use risk probability value is output.

Images