A drug screening device and method based on microsaccade and pupil nystagmus analysis

Abstract

The application discloses a drug screening device and method based on microsaccade and pupil nystagmus analysis, comprising: an information acquisition module for acquiring the multi-modal original video stream of the tester's eye, obtaining multi-modal original data; an anti-interference processing module for denoising and enhancing the multi-modal original data to obtain a high-fidelity pure image sequence; a feature extraction module for multi-dimensional feature extraction and quantitative analysis of the high-fidelity pure image sequence to obtain a standardized multi-dimensional physiological feature vector; a research and judgment module for performing drug-related risk intelligent judgment according to the standardized multi-dimensional physiological feature vector to obtain a qualitative conclusion of drug-related risk; a human-computer interaction module for providing operation control, state display and result feedback in the detection process. The screening device is non-contact, non-invasive and fast in response, has strong on-site anti-interference ability and portability, significantly reduces the missed detection rate, and improves the efficiency and accuracy of public safety detection.

Description

Technical Field

[0001] This invention belongs to the field of biometric identification and detection technology, and in particular relates to a drug screening device and method based on micro-eye movement and pupillary tremor analysis. Background Technology

[0002] Currently, rapid on-site screening of drug users mainly relies on biochemical testing methods such as urine, blood, hair, or saliva. While these methods have a certain degree of accuracy in forensic identification, they reveal many shortcomings in practical applications: the sample collection process is invasive, involves personal privacy, and can easily trigger resistance from those being tested; the testing procedures are cumbersome, and reagent reactions are time-consuming, making it difficult to meet the immediate "stop and test" requirements of high-throughput scenarios such as traffic checks, specific job physical examinations, and drug rehabilitation monitoring. In recent years, non-contact detection technologies based on ocular physiological characteristics have begun to attract attention. Existing commercial pupilmeters or eye trackers assess the nervous system state by measuring changes in pupil diameter or gaze trajectory, but their sampling frame rates are generally low, typically only capturing the macroscopic contraction and slow recovery process of the pupil.

[0003] However, existing technologies still face several key challenges: Firstly, medical research indicates that the effects of opioids or stimulant psychoactive substances on the central nervous system often manifest as microsecond-level dynamic abnormalities, such as high-frequency pupillary nystagmus above 80Hz during the light reflection recovery period and changes in microsagittal characteristics during fixation. Traditional low-frame-rate devices have sampling capabilities far below the Nyquist sampling theorem requirements, leading to the smoothing or even loss of these high-frequency pathological features, easily resulting in missed detections or misjudgments. Secondly, most existing high-precision eye trackers are limited to constant lighting environments in laboratories, are bulky, and require head fixation, making them unsuitable for the handheld portability needs of frontline law enforcement. During on-site handheld detection, ambient stray light, strong infrared reflections from the corneal surface, eyelash obstruction, eyelid drooping, and motion artifacts caused by device shaking severely interfere with the accurate positioning of the pupil edge, making it difficult for traditional image processing algorithms to separate minute nystagmus signals from a shaky image. Therefore, achieving a balance between high-frequency dynamic feature extraction and robust anti-interference capabilities in complex environments is a pressing technical challenge in the current field of drug screening technology. Summary of the Invention

[0004] This invention provides a portable drug screening method and device based on micro-eye movement and tremor analysis to solve the problems of poor timeliness and high invasiveness of existing drug detection methods, as well as the difficulty in capturing microsecond-level neuropathological features due to insufficient sampling rate and weak on-site anti-interference ability of traditional eye movement equipment, which easily leads to missed detection or misjudgment.

[0005] To address the aforementioned technical problems, this invention provides a drug screening device and method based on micro-eye movement and pupillary tremor analysis.

[0006] One of the drug screening devices based on micro-eye movement and pupillary nystagmus analysis includes: The information acquisition module is used to acquire multimodal raw video streams of the test subject's eyes to obtain multimodal raw data containing micro-eye movement trajectories and pupillary tremor signals; An anti-interference processing module, connected to the information acquisition module, is used to perform denoising and enhancement processing on the multimodal raw data to obtain a high-fidelity pure image sequence; The feature extraction module, connected to the anti-interference processing module, is used to perform multi-dimensional feature extraction and quantification analysis on the high-fidelity pure image sequence to obtain a standardized multi-dimensional physiological feature vector. The analysis module, connected to the feature extraction module, is used to perform intelligent judgment of drug-related risks based on the standardized multidimensional physiological feature vector, and obtain a qualitative conclusion on drug-related risks. The human-computer interaction module is connected to the information acquisition module, the anti-interference processing module, the feature extraction module, and the judgment module, respectively, and is used to provide operation control, status display, and result feedback during the detection process.

[0007] Preferably, the information acquisition module includes: The infrared high-speed main camera is used to continuously acquire high-speed images of the eye under both light-free and light-stimulated conditions, and obtain high frame rate video streams of micro-eye movement trajectories and pupillary tremors. A visible light auxiliary camera works synchronously with the infrared high-speed main camera to acquire facial environment images and ambient light parameters, thereby obtaining a visible light auxiliary video stream. A multi-band light source array, including an infrared supplementary light source and a visible light pulse stimulation light source, is used to provide infrared illumination during the dark adaptation phase and to emit visible light pulses during the stimulation phase to induce pupil tremor signals. ToF ranging sensors are used to monitor the distance between the device and the tester's eyes in real time and obtain distance data; The infrared high-speed main camera, the visible light auxiliary camera, the multi-band light source array, and the ToF ranging sensor are integrated into the optical imaging unit at the front end of the handheld housing, and microsecond-level time synchronization acquisition is achieved through hardware synchronization logic.

[0008] Preferably, the anti-interference processing module includes: The electronic image stabilization unit is used to calculate the global motion vector based on the background optical flow data acquired by the visible light auxiliary camera and perform inverse affine transformation compensation to obtain a stabilized eye image sequence. A spot removal unit, connected to the electronic image stabilization unit, is used to perform adaptive threshold segmentation and digital image restoration on corneal surface reflected spots in the stabilized eye image sequence to obtain a spot-removed image sequence. An occlusion processing unit, connected to the spot removal unit, is used to perform morphological filtering and random sampling consistency ellipse fitting on the eyelash occlusion and eyelid interference areas in the spot-removed image sequence to obtain a complete pupil geometric contour image sequence. An edge-preserving smoothing unit, connected to the occlusion processing unit, is used to perform bilateral filtering on the complete pupil geometric contour image sequence to obtain the high-fidelity clean image sequence.

[0009] Preferably, the occlusion processing unit includes: The candidate point acquisition subunit is used to obtain a set of candidate points from the pupil edge detection results; An iterative sampling subunit, connected to the candidate point acquisition subunit, is used to randomly select multiple non-collinear points from the candidate point set to construct an initial elliptical model. The interior point filtering subunit, connected to the iterative sampling subunit, is used to calculate the geometric distance from all candidate points to the initial elliptical model, and to mark points whose distance is less than a set threshold as interior points; The ellipse update subunit, connected to the interior point filtering subunit, is used to refit the ellipse based on all interior points and update the model parameters until the convergence condition is met, so as to obtain the optimal ellipse model after removing outliers as the complete pupil geometric contour.

[0010] Preferably, the feature extraction module includes: The geometric reconstruction unit is used to perform sub-pixel-level edge detection on the high-fidelity clean image sequence, construct the time-varying curve of pupil diameter and eye movement trajectory frame by frame, and obtain geometric morphology data; The time-domain analysis unit, connected to the geometric reconstruction unit, is used to perform differential analysis on the time-varying curve of the pupil diameter, calculate the light reflection latency, maximum contraction velocity and re-expansion hysteresis parameters, and obtain the time-domain dynamic characteristics. The frequency domain analysis unit, connected to the geometric reconstruction unit, is used to perform a fast Fourier transform on the diameter signal during the pupil recovery period, extract the tremor power spectral density, and obtain frequency domain energy characteristics. The kinematic analysis unit, connected to the geometric reconstruction unit, is used to perform kinematic modeling of the eye movement trajectory, calculate the peak velocity, acceleration and gaze stability parameters of microsaccades, and obtain motion control characteristics; The feature fusion unit is connected to the time-domain analysis unit, the frequency-domain analysis unit, and the kinematic analysis unit, respectively, and is used to fuse the time-domain dynamic features, the frequency-domain energy features, and the motion control features into the standardized multidimensional physiological feature vector.

[0011] Preferably, the frequency domain analysis unit includes: The Fast Fourier Transform (FFT) subunit is used to perform a Fast Fourier Transform on the diameter signal during the pupil recovery period to obtain the power spectral density distribution. An energy integral extraction subunit, connected to the fast Fourier transform subunit, is used to extract the flutter energy integral in the 80Hz~100Hz frequency band from the power spectral density distribution to obtain the frequency domain energy characteristics. The short-time Fourier transform subunit is used to perform short-time Fourier transform on the non-stationary diameter signal during the pupil recovery period to obtain local instantaneous frequency energy characteristics as auxiliary frequency domain features.

[0012] Preferably, the analysis module includes: The feature standardization unit is used to normalize the standardized multidimensional physiological feature vector to obtain feature input data with uniform dimensions. A multidimensional weighted evaluation unit, connected to the feature standardization unit, is used to perform weighted fusion of the deviation of each dimension in the feature input data according to preset weight coefficients to obtain a drug-related risk index. The classification and determination unit, connected to the multidimensional weighted evaluation unit, is used to compare the drug-related risk index with a preset safety threshold to obtain a qualitative conclusion on the drug-related risk and the corresponding confidence interval.

[0013] Preferably, the human-computer interaction module includes: The touch screen is used to display the device's operating status, the captured image, the preview of the area of ​​interest, the focus prompt box, the detection process status, and the qualitative conclusion of the drug-related risk. The voice prompt unit is used to play standardized testing instructions, testing progress prompts, and result warning messages; The operation control unit, including physical trigger buttons or touch control interfaces, is used to start the detection process, configure system parameters, and execute data acquisition control.

[0014] Preferably, the device further includes an embedded computing control unit, the embedded computing control unit comprising: The FPGA processor is connected to the infrared high-speed main camera, the visible light auxiliary camera, and the multi-band light source array in the information acquisition module, respectively, and is used to perform high-speed image acquisition control and multi-device timing synchronization. An ARM processor is connected to the FPGA processor and to the anti-interference processing module, the feature extraction module, and the judgment module, respectively, for executing image processing algorithms, feature calculations, and risk judgment model operations; The FPGA processor and the ARM processor are connected via an internal data bus to form a closed-loop control architecture from image acquisition and data processing to result output.

[0015] This invention also provides a drug screening method based on micro-eye movement and pupillary nystagmus analysis, comprising: According to the preset acquisition strategy and timing configuration, the infrared high-speed camera and the visible light auxiliary camera are controlled to simultaneously acquire multimodal raw video streams of the test subject's eyes, and obtain multimodal raw data including dark-adapted micro-eye movement trajectories and pupil tremor signals under light stimulation. Based on the original multimodal data, anti-interference processing and feature extraction are performed to obtain standardized multidimensional physiological feature vectors; Based on the standardized multidimensional physiological feature vector, a pre-set classification model is used to perform intelligent judgment of drug-related risks and obtain qualitative conclusions on drug-related risks.

[0016] Compared with the prior art, the present invention has the following advantages and technical effects: This invention utilizes ultra-high frame rate imaging and multimodal anti-interference algorithms to capture and quantify the neuropathological features of the human eye at the microsecond level during light reflection in real time, solving the feature loss problem caused by insufficient sampling rate and environmental interference in traditional detection methods. By constructing a multi-parameter fusion model including time domain, frequency domain, and kinematic dimensions, it achieves accurate judgment of the excitation or inhibition state of the central nervous system. Compared with biochemical detection, this invention is non-contact, non-invasive, and has a fast response speed. Compared with traditional eye trackers, this invention has strong on-site anti-interference capabilities and portability, which can effectively help front-line law enforcement personnel achieve rapid screening with "stop and test immediately", significantly reducing the false negative rate and improving the efficiency and accuracy of public safety detection. Attached Figure Description

[0017] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the device structure according to an embodiment of the present invention; Figure 2 This is a front-end lens layout diagram according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0020] Example 1 like Figure 1 As shown, this embodiment provides a drug screening device based on micro-eye movement and pupillary nystagmus analysis, comprising: The information acquisition module is used to acquire multimodal raw video streams of the test subject's eyes to obtain multimodal raw data containing micro-eye movement trajectories and pupillary tremor signals; The anti-interference processing module, connected to the information acquisition module, is used to denoise and enhance the multimodal raw data to obtain a high-fidelity, clean image sequence. The feature extraction module, connected to the anti-interference processing module, is used to perform multi-dimensional feature extraction and quantitative analysis on high-fidelity and pure image sequences to obtain standardized multi-dimensional physiological feature vectors. The analysis module, connected to the feature extraction module, is used to perform intelligent judgment of drug-related risks based on standardized multidimensional physiological feature vectors and obtain qualitative conclusions on drug-related risks. The human-computer interaction module is connected to the information acquisition module, anti-interference processing module, feature extraction module, and analysis module, respectively, and is used to provide operation control, status display, and result feedback during the detection process.

[0021] Furthermore, the information collection module includes: The infrared high-speed main camera is used to continuously acquire high-speed images of the eye under both light-free and light-stimulated conditions, and obtain high frame rate video streams of micro-eye movement trajectories and pupillary tremors. The visible light auxiliary camera works synchronously with the infrared high-speed main camera to acquire facial environmental images and ambient light parameters, thereby obtaining a visible light auxiliary video stream. A multi-band light source array, including an infrared supplementary light source and a visible light pulse stimulation light source, is used to provide infrared illumination during the dark adaptation phase and to emit visible light pulses during the stimulation phase to induce pupil tremor signals. ToF ranging sensors are used to monitor the distance between the device and the tester's eyes in real time and obtain distance data; The infrared high-speed main camera, visible light auxiliary camera, multi-band light source array and ToF ranging sensor are integrated into the optical imaging unit at the front of the handheld housing, and microsecond-level time synchronization acquisition is achieved through hardware synchronization logic.

[0022] Furthermore, the information acquisition module involved in this embodiment is used to coordinate the multimodal synchronous acquisition of a 2000fps~10000fps infrared high-speed main camera and a 600fps~1000fps visible light auxiliary camera; this module records micro-eye movement trajectories under no light stimulation conditions, captures pupil micro-contraction and 80Hz~100Hz high-frequency tremor signals under pulsed light stimulation conditions, and acquires ambient light parameters in real time, providing a high-quality raw video stream containing spatiotemporal features for subsequent processing; The information acquisition module is integrated into the optical imaging unit at the front end of the handheld housing. The optical imaging unit arranges an infrared high-speed main camera, a visible light auxiliary camera, a ToF ranging sensor and a multi-band light source array based on the coaxial imaging principle. It also reduces crosstalk between different optical paths through a physical isolation structure, thereby achieving stable multimodal eye image acquisition. Specifically, the information acquisition module includes: a time-synchronous acquisition system based on dual-spectrum imaging, integrating a 2000fps~10000fps high-speed infrared main camera with a narrowband filter and a 600fps~1000fps visible light auxiliary camera; the module controls multi-band light sources through hardware synchronization logic and executes a preset stimulus-response protocol: in the no-light-stimulation stage, infrared supplementary light is used to record the micro-eye movement trajectory under dark adaptation; in the light-stimulation stage, millisecond-level visible light pulses are triggered and the 80Hz~100Hz high-frequency tremor features during pupil micro-contraction and recovery period are continuously captured, while optical isolation technology is used to ensure that the imaging contrast is not interfered with by the stimulus light, and combined with region of interest cropping and ambient light monitoring, a high-fidelity eye video stream with spatiotemporal alignment is output; More specifically, the information acquisition module is configured as a spatiotemporal synchronous imaging system based on a dual-spectral heterogeneous sensor. Its core hardware includes an optical acquisition unit located at the front end of the handheld terminal. The main acquisition unit is a high-speed infrared camera equipped with a global shutter CMOS sensor and a specific band narrowband filter. Its sampling frame rate is set between 2000fps and 10000fps to eliminate visible light interference and capture microsecond-level eye dynamics. The auxiliary acquisition unit is a wide-angle visible light camera with a frame rate between 600fps and 1000fps to acquire facial environmental information. The information acquisition module also includes a multi-band illumination unit and a distance sensing unit. The multi-band illumination unit includes an infrared supplementary light source and a visible light pulse stimulation light source. The distance sensing unit includes a ToF ranging sensor, used to monitor the distance between the device and the subject's eyes in real time and help maintain optimal imaging depth of field. During the acquisition process, the information acquisition module executes strict microsecond-level timing logic through an FPGA or high-speed control circuit: First, in the dark adaptation reference stage, the module only turns on the infrared supplementary light, using the infrared main camera to record the spontaneous high-speed micro-scanning and fixation of the eyeball under light-sensitive conditions. The stability trajectory is then established; subsequently, the stimulus response phase begins, where the module triggers a visible light stimulus source to emit pulsed intense light with adjustable pulse width. During this phase, the infrared main camera, leveraging its spectral isolation characteristics, continuously and uninterruptedly records the rapid contraction of the pupil under intense light induction, as well as the 80Hz~100Hz high-frequency tremor signal that appears during the recovery period after the light stimulus fades. Simultaneously, the visible light auxiliary camera synchronously records the video stream of changes in ambient light intensity and relative facial displacement, ultimately outputting dual-modal raw video data containing precise timestamps, providing complete spatiotemporal data support for subsequent anti-interference processing and feature calculation.

[0023] Furthermore, the anti-interference processing module includes: The electronic image stabilization unit is used to calculate the global motion vector based on the background optical flow data acquired by the visible light auxiliary camera and perform inverse affine transformation compensation to obtain a stabilized eye image sequence. The spot removal unit, connected to the electronic image stabilization unit, is used to perform adaptive threshold segmentation and digital image inpainting on corneal surface reflective spots in the stabilized eye image sequence to obtain a spot-removed image sequence. The occlusion processing unit, connected to the spot removal unit, is used to perform morphological filtering and random sampling consistency ellipse fitting on the eyelash occlusion and eyelid interference areas in the spot-removed image sequence to obtain a complete pupil geometric contour image sequence. The edge-preserving smoothing unit, connected to the occlusion processing unit, is used to perform bilateral filtering on the complete pupil geometric contour image sequence to obtain a high-fidelity clean image sequence.

[0024] Furthermore, the occlusion processing unit includes: The candidate point acquisition subunit is used to obtain a set of candidate points from the pupil edge detection results; The iterative sampling subunit, connected to the candidate point acquisition subunit, is used to randomly select multiple non-collinear points from the candidate point set to construct an initial elliptical model. The in-point filtering subunit, connected to the iterative sampling subunit, is used to calculate the geometric distance from all candidate points to the initial elliptical model and mark points whose distance is less than a set threshold as in-points; The ellipse update sub-unit is connected to the interior point filtering sub-unit. It is used to refit the ellipse based on all interior points and update the model parameters until the convergence condition is met, so as to obtain the optimal ellipse model after removing outliers as the complete pupil geometric contour.

[0025] Furthermore, the anti-interference processing module involved in this embodiment is used to denoise and enhance the original acquired data. This module uses spectral filtering and digital image restoration algorithms to remove interference noise caused by ambient stray light, corneal reflection spots and eyelash obstruction in real time, and suppresses non-target band signals, significantly improving the image signal-to-noise ratio and providing high-fidelity clean image data for subsequent feature extraction. The anti-interference processing module includes: a high-precision denoising and restoration of the original multimodal video stream to eliminate non-target signal interference caused by physical environment and physiological characteristics; the module first uses electronic image stabilization algorithm to compensate for global displacement caused by hand shake based on auxiliary camera data, then eliminates infrared reflection spots on the corneal surface through adaptive threshold segmentation and digital image restoration technology, and further uses morphological filtering combined with random sampling consistency ellipse fitting algorithm to remove edge outliers and accurately restore the complete pupil geometric model based on local arcs under non-ideal conditions such as drooping eyelids or eyelash obstruction, thereby outputting a standardized eye image sequence with high signal-to-noise ratio and clear contour features; Furthermore, the anti-interference processing module of this embodiment incorporates a cascaded image enhancement and restoration algorithm, designed to specifically eliminate four types of core physical interference in handheld on-site detection, ensuring the authenticity of feature data. First, for the strong specular glints formed by the infrared filler array on the smooth corneal surface, the module uses brightness adaptive threshold segmentation technology to accurately locate the coordinates of the bright glints, and employs a digital image restoration algorithm (Inpainting) to intelligently fill the pupil area covered by the glints based on neighborhood texture. Second, for the incomplete pupil outline problem commonly seen in drug users due to drooping eyelids or eyelash obstruction, the module first filters out eyelash texture through morphological opening operations, and then uses the Random Sample Consistency (RANSAC) ellipse fitting algorithm to automatically identify and remove outliers in edge detection, and only inversely calculates the complete pupil geometric model based on the unobstructed local effective arc segments.

[0026] Specifically, the RANSAC ellipse fitting algorithm is a robust parameter estimation algorithm. Its core idea is to eliminate outliers through iterative sampling and fit the optimal model using only valid interior points. It is perfectly suited for pupil contour reconstruction scenarios with eyelash occlusion and eyelid interference. The specific process is as follows: (1) Input: Candidate point set P = { obtained from pupil edge detection =( , ), i=1,2,...,N}; (2) Iteration process: randomly select 5 non-collinear points from the point set P (an ellipse is a quadratic curve, and 5 parameters are required to determine a unique ellipse) to construct an initial ellipse model; calculate the geometric distance from all candidate points to the ellipse , set a distance threshold t, and mark points with <t as inliers, and the rest as outliers; re-fit the ellipse based on all inliers and update the model parameters; repeat the above sampling-verification-update steps until the number of iterations reaches the upper limit or the number of inliers and model residual meet the convergence condition.

[0027] (3) Output: the optimal ellipse model after removing outliers, which is the complete geometric contour of the pupil.

[0028] Core formula for ellipse fitting: (1) General conic equation of an ellipse; Any ellipse in a plane can be expressed in the form of a quadratic curve: where A, B, C, D, E, F are ellipse parameters, satisfying the constraint condition B²-4AC<0 (ensuring the curve is an ellipse, not a hyperbola / parabola).

[0029] Calculation of geometric distance from a point to an ellipse For a candidate point =( , ), the geometric distance from the point to the ellipse is calculated by the residual formula: Set the threshold t, when | | < t, the point is determined as an inlier and participates in the final ellipse fitting.

[0030] (3) Least squares optimization; Based on the inlier set , the optimal ellipse parameters are solved by minimizing the sum of squared residuals: The parameter vector is solved by the matrix pseudo-inverse method = to obtain the optimal ellipse model.

[0031] Meanwhile, to eliminate motion artifacts and high-frequency jitter that are unavoidable during handheld operation, the module uses background optical flow data acquired by the auxiliary camera as a reference frame to calculate the global motion vector and perform electronic image stabilization (EIS) and inverse affine transformation compensation to eliminate rigid displacement of the image in the temporal domain. Finally, to address abrupt changes in ambient stray light and high-gain thermal noise from the sensor, the module applies edge-preserving smoothing algorithms such as bilateral filtering to suppress background salt-and-pepper noise while maximizing the preservation of grayscale gradient sharpness at the pupil edge, thereby outputting a standardized eye video sequence with high signal-to-noise ratio, clear contours, and stable background.

[0032] Furthermore, the feature extraction module includes: The geometric reconstruction unit is used to perform sub-pixel-level edge detection on high-fidelity clean image sequences, and to construct the time-varying curve of pupil diameter and eye movement trajectory frame by frame to obtain geometric morphology data. The time-domain analysis unit, connected to the geometric reconstruction unit, is used to perform differential analysis on the time-varying curve of pupil diameter, calculate the light reflection latency, maximum contraction velocity and re-expansion hysteresis parameters, and obtain the time-domain dynamic characteristics. The frequency domain analysis unit, connected to the geometric reconstruction unit, is used to perform a fast Fourier transform on the diameter signal during the pupil recovery period, extract the tremor power spectral density, and obtain frequency domain energy characteristics. The kinematic analysis unit, connected to the geometric reconstruction unit, is used to perform kinematic modeling of eye movement trajectories, calculate the peak velocity, acceleration, and gaze stability parameters of microsaccades, and obtain motion control characteristics. The feature fusion unit is connected to the time domain analysis unit, the frequency domain analysis unit, and the kinematic analysis unit, respectively, and is used to fuse time domain dynamic features, frequency domain energy features, and motion control features into a standardized multidimensional physiological feature vector.

[0033] Furthermore, the frequency domain analysis unit includes: The Fast Fourier Transform (FFT) subunit is used to perform a Fast Fourier Transform on the diameter signal during the pupil recovery period to obtain the power spectral density distribution. The energy integral extraction subunit, connected to the fast Fourier transform subunit, is used to extract the flutter energy integral in the 80Hz~100Hz frequency band from the power spectral density distribution to obtain the frequency domain energy characteristics. The short-time Fourier transform subunit is used to perform short-time Fourier transform on the non-stationary diameter signal during the pupil recovery period to obtain local instantaneous frequency energy characteristics as auxiliary frequency domain features.

[0034] Furthermore, the feature extraction module involved in this embodiment is used to perform feature extraction and quantitative analysis on image sequences after anti-interference. This module uses morphological and frequency domain transformation algorithms to solve the pupillary contraction latency and dynamic recovery features under light stimulation, extracts the tremor energy spectrum in the 80Hz~100Hz frequency band, and calculates the saccade rate and fixation stability of micro-eye movements. Finally, it transforms the above multi-dimensional spatiotemporal information into standardized feature vectors and transmits them to the judgment module for decision-making. The feature extraction module includes: a feature extraction module for performing the transformation and extraction from the original video stream to a high-dimensional physiological feature vector; the module first constructs the pupil diameter time-varying curve and eye movement trajectory based on a sub-pixel level geometric measurement algorithm, then calculates the light reflection latency, maximum contraction velocity and re-expansion hysteresis parameters in the time domain, and extracts the tremor power spectral density (PSD) of the pupil recovery period in the 80Hz-100Hz frequency band through Fast Fourier Transform (FFT) in the frequency domain, and calculates the peak velocity, the master order relationship of acceleration and fixation stability (BCEA) of microsagittal movement by combining the eye kinematic model, and finally outputs a quantitative data vector containing neural reflex features and motor control features to the judgment module; Furthermore, the feature extraction module in this embodiment is configured to perform deep feature mining and parameter calculation on the image sequence after anti-interference processing. Its processing flow spans four dimensions: geometry, time domain, frequency domain, and kinematics. The module first uses sub-pixel edge detection technology to construct a high-precision pupil diameter change curve and eye gaze trajectory frame by frame from the video stream. Based on this basic geometric data, on the one hand, it performs time-domain differential analysis on the diameter curve to calculate dynamic response indicators such as light reflection latency, contraction speed, and re-expansion hysteresis. It further performs frequency domain transformation on the recovery period data to extract and quantify the tremor energy spectral density unique to the 80Hz~100Hz frequency band. On the other hand, the module simultaneously performs kinematic modeling on the eye gaze trajectory, calculates the peak velocity, acceleration, and principal order relationship of micro-eye movements, and calculates the gaze stability score by combining spatial scattering. Finally, it integrates the geometric morphology data, dynamic response parameters, high-frequency tremor energy, and motion control features extracted from different physical dimensions into a standardized multi-dimensional physiological feature vector, providing quantitative numerical basis for subsequent drug-related assessment.

[0035] Furthermore, the analysis module includes: The feature standardization unit is used to normalize the standardized multidimensional physiological feature vector to obtain feature input data with uniform dimensions. The multidimensional weighted evaluation unit, connected to the feature standardization unit, is used to weight and fuse the deviation of each dimension in the feature input data according to the preset weight coefficients to obtain the drug-related risk index. The classification and judgment unit, connected to the multidimensional weighted evaluation unit, is used to compare the drug-related risk index with a preset safety threshold to obtain a qualitative conclusion on the drug-related risk and the corresponding confidence interval.

[0036] Furthermore, the judgment module involved in this embodiment is used to perform intelligent judgment of drug-related risks based on multi-dimensional feature vectors. The module uses a preset classification model or multi-parameter threshold logic to comprehensively evaluate the high-frequency pupillary tremor energy, light reflection dynamic parameters and eye movement stability, calculate the matching degree between the test subject's data and the drug-related physiological model, and finally output a qualitative conclusion on whether or not the subject is involved in drugs. The judgment module, configured as the system's decision-making core, receives the multi-dimensional feature vectors output by the feature extraction module and performs intelligent judgment using a pre-built database of drug-related physiological characteristics and a multi-parameter fusion classification model. This module first performs a weighted evaluation of the input geometric dynamic parameters, frequency domain energy values, and kinematic indices, calculating the test subject's deviation in each independent feature dimension through logistic regression analysis. Subsequently, the module executes joint inference logic: if a significant 80Hz~100Hz high-frequency tremor energy peak is detected during the pupillary recovery period, accompanied by abnormal deviations in the micro-eye movement master sequence or nonlinear prolongation of the light reflex latency, the central nervous system is determined to be in a state of excitation or inhibition disorder. Finally, the module integrates the weighted scores of each dimension and calculates a quantitative "Drug Risk Index (DRI)" and its corresponding confidence interval using a weighted fusion formula. The core calculation formula is as follows: Where n is the total number of feature dimensions involved in the analysis. The preset weight coefficients corresponding to the i-th feature dimension (satisfying) ), The deviation score is the standardized value for the i-th feature dimension. When the index exceeds the preset safety threshold, a qualitative conclusion of "suspected drug involvement" is generated, and the intuitive judgment result and main abnormal features are output on the display screen.

[0037] Furthermore, the human-computer interaction module includes: The touch screen is used to display the device's operating status, captured images, previews of areas of interest, focus prompts, detection process status, and qualitative conclusions regarding drug-related risks. The voice prompt unit is used to play standardized testing instructions, testing progress prompts, and result warning messages; The operation control unit, including physical trigger buttons or touch control interfaces, is used to start the detection process, configure system parameters, and execute data acquisition control.

[0038] Furthermore, the human-computer interaction module involved in this embodiment is used to provide operation control, status display and result feedback during the detection process; the module includes a touch screen, operation buttons and a voice prompt unit, wherein the touch screen is used to display the acquisition status, detection process and detection results, the speaker is used to play detection instructions and prompt information, and the operation buttons are used to trigger the detection process and system control, thereby realizing the interaction between the detection device and the operator and the person being tested; The human-machine interaction module includes a touch screen, a voice prompt unit, and an operation control unit; wherein the touch screen is used to display the equipment operating status, testing process, and testing results; the voice prompt unit is used to play standardized testing instructions to the tested personnel and indicate the testing progress; the operation control unit includes physical buttons or a touch control interface, used to start the testing process, configure parameters, and control the system, thereby realizing real-time interaction between the testing device and the operator; More specifically, the human-computer interaction module provides operation control, status display, guidance prompts, and result feedback during the testing process. This module includes a touch display unit, a voice prompt unit, and an operation control unit located at the back of the handheld terminal. The touch display unit displays the acquired image, region of interest preview, focus prompt box, testing process status, and final testing result in real time. The voice prompt unit plays standardized voice guidance information, testing progress prompts, and result warnings. The operation control unit includes trigger buttons or a touch control interface for starting the testing process, executing parameter settings, and performing system control, thereby realizing interaction between the testing device, the operator, and the tested individual.

[0039] Furthermore, the device also includes an embedded computing control unit, which includes: The FPGA processor is connected to the infrared high-speed main camera, the visible light auxiliary camera, and the multi-band light source array in the information acquisition module, respectively, and is used to perform high-speed image acquisition control and multi-device timing synchronization. An ARM processor is connected to an FPGA processor, and is also connected to an anti-interference processing module, a feature extraction module, and an analysis module, respectively, to execute image processing algorithms, feature calculations, and risk assessment model operations. The FPGA processor and the ARM processor are connected through an internal data bus to form a closed-loop control architecture from image acquisition and data processing to result output.

[0040] Furthermore, the device usage steps of this embodiment include: S0. System initialization and self-test steps: Start the above portable drug screening device. The embedded main control unit completes the power-on initialization and self-test of the infrared high-speed camera, visible light auxiliary camera, multi-band light source and ranging sensor; loads the preset drug detection protocol, and automatically calibrates the gain parameters of the infrared camera and the pulse intensity of the visible light stimulus source according to the ambient light intensity. S1. Acquisition Strategy and Timing Configuration Steps: Before the detection process is locked, the main control unit sets the working timing of the information acquisition module, specifically including: configuring the sampling frame rate of the infrared main camera to 2000fps~10000fps, setting the single pulse duration and trigger delay time of the visible light stimulus source, and allocating storage space in the high-speed cache area to form a high-throughput acquisition strategy that matches the entire process of "dark adaptation-light stimulation-recovery period"; S2. Subject guidance and intelligent alignment steps: Standardized voice commands are played through the speaker of the human-computer interaction module to guide the subject to look at the front of the device. At the same time, the ToF ranging sensor is activated to monitor the distance between the device and the eye in real time. When the distance is within the effective depth of field and the auxiliary camera detects that the eye ROI is located in the center of the image, a green focusing frame is displayed on the screen and the operator is automatically triggered or prompted to press the operation key to enter the acquisition state. S3. Dark-adapted steady-state acquisition steps: Under dark background conditions where no visible light stimulus is triggered, only the infrared supplementary light array is turned on to control the infrared main camera to acquire the first stage of eye image sequence, which is used to capture the microsaccade trajectory and static pupil diameter baseline value of the subject in the absence of light stimulus, as a zero-point reference for subsequent dynamic analysis. S4. Transient Stimulation and High-Frequency Dynamic Acquisition Steps: After the baseline acquisition is completed, the main control unit drives the visible light stimulation source to emit millisecond-level strong pulse light, and synchronously controls the infrared main camera to continuously and uninterruptedly acquire the second-stage eye image sequence during the moment of stimulation triggering, the duration of stimulation, and the recovery period after stimulation fades, so as to fully record the rapid pupil contraction process and the 80Hz-100Hz high-frequency tremor (drifts) signal that appears during the recovery period; S5. Anti-interference calculation and drug-related assessment steps: The original video streams obtained in steps S3 and S4 are input to the anti-interference processing module and the feature extraction module. First, electronic image stabilization, spot removal, and RANSAC-based pupil geometry reconstruction are performed. Then, time-domain differential and frequency-domain FFT algorithms are used to extract feature parameters such as light reflection latency, re-dilation hysteresis time, and tremor energy spectral density. Finally, the multi-dimensional feature vector is input to the machine learning-based assessment model, which outputs the drug-related risk index, confidence interval, and corresponding abnormal feature type of the tested person.

[0041] Therefore, this embodiment addresses the problem that existing low-frame-rate devices cannot capture high-frequency pathological features. It employs a high-speed infrared main camera (2000fps~10000fps) and a visible light auxiliary camera (600fps~1000fps) for collaborative acquisition, combined with a pulsed light stimulation protocol. This achieves, for the first time, complete recording and quantitative analysis of 80Hz~100Hz high-frequency pupillary tremor signals and micro-eye movement trajectories during the light reflection recovery period, effectively avoiding the loss of high-frequency features and significantly reducing the false negative rate. To address the strong interference problems in on-site handheld detection, such as environmental stray light, corneal reflection spots, eyelash obstruction, and equipment shake, this embodiment uses anti-interference processing modules such as spectral filtering, electronic image stabilization, digital image restoration, and ellipse fitting based on Random Sample Consistency (RANSAC). This robustly reconstructs the complete pupillary geometric contour, eliminating the influence of various physical noises on feature extraction, thus extending high-precision screening from laboratory environments to portable handheld law enforcement scenarios. Ultimately, by constructing a multi-dimensional feature vector that integrates time domain, frequency domain, and kinematic dimensions, along with an intelligent judgment module, a rapid and non-invasive determination of the excitation or inhibition state of the central nervous system was achieved. This enabled a rapid on-site screening effect of "stop and test immediately," significantly improving the efficiency and accuracy of public safety detection.

[0042] As a preferred implementation method, such as Figure 1 As shown, this embodiment provides a handheld optoelectronic integrated detection terminal for rapidly collecting micro-eye movement and pupil dynamic change information of the subject's eyes on-site, and intelligently assessing drug-related risks. The device adopts a portable handheld design and mainly includes: a handheld shell assembly, a front-end optical acquisition assembly, a rear-end human-computer interaction assembly, and an embedded computing control unit. These components are connected and work collaboratively via an internal data bus to form an integrated detection device.

[0043] Handheld housing structure: The handheld housing assembly adopts an ergonomic gun-style design, including a main housing and a grip handle connected to it. The main housing houses the optical acquisition components and embedded computing unit, and its interior is coated with an anti-reflective layer to reduce the impact of internal stray light reflection on image quality. The grip handle integrates a power module, preferably a rechargeable lithium battery pack, to provide stable power to the high-speed camera, light source system, and computing unit.

[0044] In this embodiment, a physical trigger button is provided at the front end of the handle for initiating the detection process. In a preferred embodiment, the trigger button may have a two-stage structure to support the operation logic of half-pressing for focusing and full-pressing for acquisition; Front-end optical acquisition components: like Figure 2As shown, the front panel of the main housing integrates an optical acquisition component, which arranges multi-dimensional sensing units based on the coaxial imaging principle. The optical acquisition component includes: Infrared high-speed main camera: Located at the geometric center of the panel, it adopts a global shutter CMOS sensor and is configured with a sampling frame rate of 2000fps to 10000fps; a narrow band filter with a center wavelength of 850nm is mounted in front of the lens to ensure that only the pupil image in the infrared band is received during visible light stimulation, thus achieving spectral physical isolation; Multi-band illumination unit: The multi-band illumination unit is arranged around the infrared high-speed main camera and includes an infrared supplementary light array and a visible light stimulation source. The infrared supplementary light array is preferably a ring-shaped infrared LED array, used to provide a uniform and stable dark-field illumination environment during the dark adaptation acquisition phase; the visible light stimulation source is located above or adjacent to the main camera, used to emit pulsed stimulation light according to preset pulse width and intensity parameters to induce rapid pupil contraction and the dynamic response during the subsequent recovery process. Auxiliary sensing unit: A ToF laser rangefinder is integrated on the side of the panel to monitor the vertical distance between the device and the eye in real time; and a visible light auxiliary camera is set up next to the main camera to capture facial environmental information and assist in electronic image stabilization; Backend human-computer interaction components: The device is equipped with a human-computer interaction component at the rear end, which is used to complete the operation control and result display during the detection process.

[0045] This component includes: Touchscreen display: The touchscreen display is used to show in real time: the captured image, the eye ROI area, the focus indicator, the detection status information, and the final detection result. After the detection is completed, the display can also show the drug risk level and abnormal characteristics.

[0046] Voice prompt unit: A speaker is installed on the side of the device to play voice prompts during the detection process, such as "Please keep your eyes on the device", "Detection started", "Detection completed". When an abnormal risk is detected, an alarm prompt tone can also be output.

[0047] Operation control unit: The operation control unit includes physical buttons or touch control interface, used to: start the detection process, perform data acquisition operations and set parameters.

[0048] Embedded computing control unit: The device also integrates an embedded computing control unit.

[0049] In this embodiment, the unit includes: an FPGA processor and an ARM processor. Wherein: The FPGA is mainly responsible for high-speed image acquisition control and timing synchronization of multiple devices; The ARM processor is responsible for executing image processing algorithms, feature calculations, and risk assessment models.

[0050] The components are connected through an internal data bus, thus realizing a complete closed-loop detection system from image acquisition and data processing to detection result output.

[0051] Example 2 like Figure 3 As shown, based on the same inventive concept, this embodiment also provides a drug screening method based on micro-eye movement and pupillary nystagmus analysis, including: According to the preset acquisition strategy and timing configuration, the infrared high-speed camera and the visible light auxiliary camera are controlled to simultaneously acquire multimodal raw video streams of the test subject's eyes, and obtain multimodal raw data including dark-adapted micro-eye movement trajectories and pupil tremor signals under light stimulation. Based on the original multimodal data, anti-interference processing and feature extraction are performed to obtain standardized multidimensional physiological feature vectors; Based on standardized multidimensional physiological feature vectors, a pre-set classification model is used to perform intelligent judgment of drug-related risks and obtain qualitative conclusions on drug-related risks.

[0052] Furthermore, the drug screening method in this embodiment specifically includes the following steps: S0. Data Acquisition and Preprocessing Steps: The information acquisition module is used to acquire dual-modal eye video data, including high-speed infrared images and visible light-assisted images, and the acquired data is subjected to spatiotemporal synchronization, scale normalization, region of interest cropping, and standardization preprocessing. S1. Anti-interference and feature extraction steps: Anti-interference processing is performed on the preprocessed eye video sequence to eliminate the effects of light spots, occlusion and hand shaking, and the geometric contour of the pupil is reconstructed; on this basis, the pupil diameter change curve and eye gaze trajectory are constructed, and micro-eye movement features and pupil nystagmus features are extracted. S2. Drug-related risk modeling steps: Input the extracted multidimensional physiological features into the drug-related risk assessment model, and establish the judgment relationship from physiological features to drug-related risk probability through feature standardization, weighted fusion and nonlinear mapping; S3. Model Training and Evaluation Steps: The evaluation model is trained using a combination of parameter optimization and cross-validation, and its performance is evaluated using metrics such as accuracy, F1 score, AUC, and confusion matrix. S4. Robustness Validation and Update Steps: The robustness of the model is validated using samples related to different environmental conditions and different types of psychoactive substances. The model parameters are updated and optimized based on the misjudged samples to improve the system's adaptability in real-world application scenarios. S5. Results Output and Interactive Feedback Steps: Output the detection results and risk warning information through the human-computer interaction module, and combine voice prompts, focus guidance and visual interface to assist in completing the detection process and result interpretation; As a preferred implementation method, the drug screening method based on the above-mentioned device is as follows: Figure 3 As shown, this embodiment details the specific processing flow and algorithm implementation for drug screening using the aforementioned device; specifically including: S0 Data Acquisition and Preprocessing Module: After the device is activated, it automatically triggers the acquisition process by locking onto an effective distance of 5-10cm in real time using Time-of-Flight (ToF) ranging. First, it acquires 2 seconds of dark-adapted steady-state data under infrared illumination alone. Then, it triggers a high-intensity visible light pulse and continuously acquires 4 seconds of recovery period data using the main infrared camera at ≥2000fps. The acquisition sequence is combined with background optical flow acquired by the auxiliary camera for electronic image stabilization (EIS) to eliminate global displacement interference caused by handheld operation, thereby establishing a high spatiotemporal resolution initial sample library and providing high-quality input for subsequent feature extraction. S1 Multidimensional Signal Repair and Time-Frequency Feature Fusion Module: The multidimensional signal restoration and time-frequency feature fusion module includes an image restoration unit and a signal analysis unit, used for anti-interference feature reconstruction and multi-scale time-frequency feature extraction of the acquired sequence. In the image restoration unit, corneal reflection spots are first precisely located in each frame. Candidate spot masks are generated through adaptive threshold segmentation. For areas potentially obscured by eyelashes or eyelids, outliers are removed using Random Sample Consensus (RANSAC), and only inliers with a confidence level higher than 85% are selected for weighted least-squares fitting to achieve accurate reconstruction of the spot's geometric position. For missing or obscured spot areas, the Inpainting algorithm, combined with local brightness smoothing and neighborhood texture matching, is used for completion. Simultaneously, spot trajectory smoothing is performed between frames to reduce continuity errors caused by hand shake. Subsequently, in the signal analysis unit, the pupil diameter curve D(t) is calculated from the pupil contour output by the image restoration. Multi-order de-trending processing is used to eliminate macroscopic baseline drift, while an adaptive filter is used to suppress low-frequency environmental noise. In the multi-scale time-frequency analysis, the power spectral density (PSD) is first calculated using Fast Fourier Transform (FFT) combined with the Hanning window to extract the energy integral (MEI) of high-frequency tremors in the recovery period (80Hz-100Hz). For non-stationary components, Short-Time Fourier Transform (STFT) and Wavelet Packet Decomposition are further used to capture local instantaneous frequency energy features. Simultaneously, temporal dynamics analysis is performed to calculate indices such as light reflection latency, maximum contraction velocity, microsaccade sequence, and instantaneous acceleration. The high-frequency features in the frequency domain, temporal dynamics indices, and microsaccade sequence are integrated in multiple dimensions to form a high-dimensional multimodal physiological feature vector. Finally, time-domain, frequency-domain, and spatial dynamic features are fused through multi-layer embedding, and a regularized decoupling strategy is used to suppress redundant information, ensuring that the output features are independent and complete in each dimension. This provides high-quality, interference-resistant, multi-scale, and multimodal feature encoding for subsequent decision-making models. S2 Multimodal High-Dimensional Decision Classification Module: The multidimensional feature vector output by S1 is standardized and mapped to a high-dimensional space, then input into a classifier based on Support Vector Machine (SVM) or Gradient Boosting Decision Tree (GBDT). The classifier combines kernel function mapping and joint decision logic, integrating frequency domain energy integral and time domain dynamic features to achieve abnormal state probability output, fully utilizing multimodal information to enhance decision accuracy.

[0053] S3 Model Training and Performance Evaluation Module: A strategy combining grid search and cross-validation is adopted, and gradient descent is used to iteratively optimize the hyperparameters and decision boundaries of the classification model. At the same time, multi-dimensional statistical indicators such as classification accuracy, F1 score, area under the ROC curve (AUC), and confusion matrix are used to evaluate the model performance on the internal test set to ensure that the model can maintain high sensitivity and high specificity under different pupil datums. S4 Robustness Verification and Adaptive Update Module: The robustness of the model was validated using external test samples collected from different lighting environments and different types of psychoactive substance abusers. The model's classification differences on marginal samples were analyzed in detail, and a feedback mechanism was established based on misjudged samples to update and fine-tune the model weights online to adapt to the variability of actual law enforcement scenarios. S5 Interactive Guidance and Visual Feedback Logic Setting Module: The human-computer interaction module provides real-time feedback on the focus status based on the collected data. During the analysis phase, it draws the dynamic pupil curve D(t) on the screen in real time and outputs graded visual prompts based on the classification probability. The normal state is displayed in green as "No abnormality found". When the abnormal probability exceeds the preset threshold, it displays in red as "Abnormal prompt". At the same time, the key feature area that led to the judgment is highlighted, and a prompt sound is played through the speaker. This realizes a closed-loop operation of data collection, judgment and visualization, providing operators with intuitive decision-making basis. The drug screening method based on micro-eye movement and pupillary tremor analysis provided in this embodiment has all the advantages of the drug screening device based on micro-eye movement and pupillary tremor analysis provided in Embodiment 1.

[0054] Example 3 This embodiment also discloses a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in Embodiment 1.

[0055] Example 4 This embodiment also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0056] Example 5 This embodiment also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0057] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A drug screening device based on micro-eye movement and pupillary nystagmus analysis, characterized in that, include: The information acquisition module is used to acquire multimodal raw video streams of the test subject's eyes to obtain multimodal raw data containing micro-eye movement trajectories and pupillary tremor signals; An anti-interference processing module, connected to the information acquisition module, is used to perform denoising and enhancement processing on the multimodal raw data to obtain a high-fidelity pure image sequence; The feature extraction module, connected to the anti-interference processing module, is used to perform multi-dimensional feature extraction and quantification analysis on the high-fidelity pure image sequence to obtain a standardized multi-dimensional physiological feature vector. The analysis module, connected to the feature extraction module, is used to perform intelligent judgment of drug-related risks based on the standardized multidimensional physiological feature vector, and obtain a qualitative conclusion on drug-related risks. The human-computer interaction module is connected to the information acquisition module, the anti-interference processing module, the feature extraction module, and the judgment module, respectively, and is used to provide operation control, status display, and result feedback during the detection process.

2. The apparatus according to claim 1, characterized in that, The information collection module includes: The infrared high-speed main camera is used to continuously acquire high-speed images of the eye under both light-free and light-stimulated conditions, and obtain high frame rate video streams of micro-eye movement trajectories and pupillary tremors. A visible light auxiliary camera works synchronously with the infrared high-speed main camera to acquire facial environment images and ambient light parameters, thereby obtaining a visible light auxiliary video stream. A multi-band light source array, including an infrared supplementary light source and a visible light pulse stimulation light source, is used to provide infrared illumination during the dark adaptation phase and to emit visible light pulses during the stimulation phase to induce pupil tremor signals. ToF ranging sensors are used to monitor the distance between the device and the tester's eyes in real time and obtain distance data; The infrared high-speed main camera, the visible light auxiliary camera, the multi-band light source array, and the ToF ranging sensor are integrated into the optical imaging unit at the front end of the handheld housing, and microsecond-level time synchronization acquisition is achieved through hardware synchronization logic.

3. The apparatus according to claim 1, characterized in that, The anti-interference processing module includes: The electronic image stabilization unit is used to calculate the global motion vector based on the background optical flow data acquired by the visible light auxiliary camera and perform inverse affine transformation compensation to obtain a stabilized eye image sequence. A spot removal unit, connected to the electronic image stabilization unit, is used to perform adaptive threshold segmentation and digital image restoration on corneal surface reflected spots in the stabilized eye image sequence to obtain a spot-removed image sequence. An occlusion processing unit, connected to the spot removal unit, is used to perform morphological filtering and random sampling consistency ellipse fitting on the eyelash occlusion and eyelid interference areas in the spot-removed image sequence to obtain a complete pupil geometric contour image sequence. An edge-preserving smoothing unit, connected to the occlusion processing unit, is used to perform bilateral filtering on the complete pupil geometric contour image sequence to obtain the high-fidelity clean image sequence.

4. The apparatus according to claim 3, characterized in that, The occlusion processing unit includes: The candidate point acquisition subunit is used to obtain a set of candidate points from the pupil edge detection results; An iterative sampling subunit, connected to the candidate point acquisition subunit, is used to randomly select multiple non-collinear points from the candidate point set to construct an initial elliptical model. The interior point filtering subunit, connected to the iterative sampling subunit, is used to calculate the geometric distance from all candidate points to the initial elliptical model, and to mark points whose distance is less than a set threshold as interior points; The ellipse update subunit, connected to the interior point filtering subunit, is used to refit the ellipse based on all interior points and update the model parameters until the convergence condition is met, so as to obtain the optimal ellipse model after removing outliers as the complete pupil geometric contour.

5. The apparatus according to claim 1, characterized in that, The feature extraction module includes: The geometric reconstruction unit is used to perform sub-pixel-level edge detection on the high-fidelity clean image sequence, construct the time-varying curve of pupil diameter and eye movement trajectory frame by frame, and obtain geometric morphology data; The time-domain analysis unit, connected to the geometric reconstruction unit, is used to perform differential analysis on the time-varying curve of the pupil diameter, calculate the light reflection latency, maximum contraction velocity and re-expansion hysteresis parameters, and obtain the time-domain dynamic characteristics. The frequency domain analysis unit, connected to the geometric reconstruction unit, is used to perform a fast Fourier transform on the diameter signal during the pupil recovery period, extract the tremor power spectral density, and obtain frequency domain energy characteristics. The kinematic analysis unit, connected to the geometric reconstruction unit, is used to perform kinematic modeling of the eye movement trajectory, calculate the peak velocity, acceleration and gaze stability parameters of microsaccades, and obtain motion control characteristics; The feature fusion unit is connected to the time-domain analysis unit, the frequency-domain analysis unit, and the kinematic analysis unit, respectively, and is used to fuse the time-domain dynamic features, the frequency-domain energy features, and the motion control features into the standardized multidimensional physiological feature vector.

6. The apparatus according to claim 5, characterized in that, The frequency domain analysis unit includes: The Fast Fourier Transform (FFT) subunit is used to perform a Fast Fourier Transform on the diameter signal during the pupil recovery period to obtain the power spectral density distribution. An energy integral extraction subunit, connected to the fast Fourier transform subunit, is used to extract the flutter energy integral in the 80Hz~100Hz frequency band from the power spectral density distribution to obtain the frequency domain energy characteristics. The short-time Fourier transform subunit is used to perform short-time Fourier transform on the non-stationary diameter signal during the pupil recovery period to obtain local instantaneous frequency energy characteristics as auxiliary frequency domain features.

7. The apparatus according to claim 1, characterized in that, The analysis module includes: The feature standardization unit is used to normalize the standardized multidimensional physiological feature vector to obtain feature input data with uniform dimensions. A multidimensional weighted evaluation unit, connected to the feature standardization unit, is used to perform weighted fusion of the deviation of each dimension in the feature input data according to preset weight coefficients to obtain a drug-related risk index. The classification and determination unit, connected to the multidimensional weighted evaluation unit, is used to compare the drug-related risk index with a preset safety threshold to obtain a qualitative conclusion on the drug-related risk and the corresponding confidence interval.

8. The apparatus according to claim 1, characterized in that, The human-computer interaction module includes: The touch screen is used to display the device's operating status, the captured image, the preview of the area of ​​interest, the focus prompt box, the detection process status, and the qualitative conclusion of the drug-related risk. The voice prompt unit is used to play standardized testing instructions, testing progress prompts, and result warning messages; The operation control unit, including physical trigger buttons or touch control interfaces, is used to start the detection process, configure system parameters, and execute data acquisition control.

9. The apparatus according to claim 1, characterized in that, The device further includes an embedded computing control unit, the embedded computing control unit comprising: The FPGA processor is connected to the infrared high-speed main camera, the visible light auxiliary camera, and the multi-band light source array in the information acquisition module, respectively, and is used to perform high-speed image acquisition control and multi-device timing synchronization. An ARM processor is connected to the FPGA processor and to the anti-interference processing module, the feature extraction module, and the judgment module, respectively, for executing image processing algorithms, feature calculations, and risk judgment model operations; The FPGA processor and the ARM processor are connected via an internal data bus to form a closed-loop control architecture from image acquisition and data processing to result output.

10. A drug screening method based on micro-eye movement and pupillary nystagmus analysis, characterized in that, include: According to the preset acquisition strategy and timing configuration, the infrared high-speed camera and the visible light auxiliary camera are controlled to simultaneously acquire multimodal raw video streams of the test subject's eyes, and obtain multimodal raw data including dark-adapted micro-eye movement trajectories and pupil tremor signals under light stimulation. Based on the original multimodal data, anti-interference processing and feature extraction are performed to obtain standardized multidimensional physiological feature vectors; Based on the standardized multidimensional physiological feature vector, a pre-set classification model is used to perform intelligent judgment of drug-related risks and obtain qualitative conclusions on drug-related risks.

Images