Non-stationary pupil signal blind source separation method based on Gamma unit impulse response model

Abstract

The invention discloses a non-stationary pupil signal blind source separation method based on a Gamma unit impulse response model. Compared with the prior art, the method has the advantages that the ecological effectiveness of clinical evaluation is improved; high-precision verification with zero hardware cost is realized; and the problem of time sequence asynchronization of physiological signals is solved.

Description

Technical Field

[0001] This invention relates to a signal processing method for separating the emotional component of the autonomic nervous system and the light reflection component of the visual nervous system from the total pupillary change signal in the objective quantitative assessment of mental and psychological states (such as cognitive load, anxiety, and depression), using multi-source heterogeneous data fusion technology. It belongs to the fields of psychophysiological monitoring, computer data processing technology, and human-computer interaction. Background Technology

[0002] In current non-contact psychological assessment or affective computing technologies, changes in pupil diameter are considered a key physiological indicator reflecting a user's cognitive load and emotional arousal. Existing technologies typically acquire eye images using infrared eye trackers and extract pupil diameter sequences using image segmentation algorithms, but these methods have the following drawbacks: Defect 1: Physiological signal aliasing The pupil diameter is controlled by two neural pathways: the sympathetic nervous system (slight dilation due to emotion / cognition) and the parasympathetic nervous system (strong constriction due to light exposure). In real-world digital therapy or psychological testing scenarios (such as browsing the web or watching videos), the brightness of the screen content changes dynamically. A sudden brightening of the screen (such as a white pop-up window) will trigger a strong pupillary light reflex.

[0003] Defect 2: Excessive artifact amplitude The contraction amplitude of physiological PLR (which can reach 1-2 mm) is much greater than the expansion amplitude caused by psychological stress (usually <0.5 mm). If this distinction is not made, existing algorithms will misjudge "contraction caused by being flashed by light" as "a drastic fluctuation in psychological state," leading to serious distortion of the assessment results.

[0004] Defect 3: Hardware dependency or scenario limitation To avoid light interference, current medical solutions typically require patients to look at a gray background screen under constant lighting (sacrificing ecological validity) or to wear additional light sensors (increasing hardware costs and system complexity). Summary of the Invention

[0005] The purpose of this invention is to solve the technical problems of effectively identifying and eliminating non-emotional pupillary light reflection artifacts caused by screen light stimulation in complex visual interaction scenarios where no external light sensor is required and the brightness of the screen content changes dynamically, as well as the technical problems of noise residue caused by the inability to accurately synchronize the light signals collected by external sensors and the pupil signals collected by eye trackers in terms of timing, thereby extracting pure pupillary physiological characteristic parameters that are only related to psychological emotions.

[0006] To achieve the above objectives, the present invention discloses a method for blind source separation of non-stationary pupil signals based on a Gamma unit impulse response model, characterized by comprising the following steps: Before the test began, a completely black screen and a completely white screen were played, and the maximum pupil diameter of the subjects was recorded respectively. and minimum pupil diameter ,like If the ambient light interference is too great, the process will be terminated; otherwise, proceed to the next step. Raw pupil diameter data is collected using an eye tracker, while the current frame of the image is captured in the graphics card's video memory using the operating system's hook function. A 10-degree visual field region centered on the fixation point is defined as the effective luminance stimulus region, and the weighted average grayscale value of the pixels within this region is calculated as the input signal. ; Construct a linear time-invariant system model and use the Gamma probability density function as the system's unit impulse response. , ,in: For shape parameters; It is the time difference between the retina receiving light and the pupillary muscle beginning to contract; it is a time constant. Input signal and Discrete convolution operations are performed to obtain the predicted light reflection components, and the optimal scaling factor is solved using the least squares method. Subtracting the predicted light reflection component from the original pupil data yields the emotion-based pupil data after artifact removal.

[0007] Preferably, the playback time for both the pure black screen and the pure white screen is 5 seconds.

[0008] Preferably, when collecting raw pupil diameter data using an eye tracker, the sampling frequency is not less than 60Hz.

[0009] Preferably, when acquiring raw pupil diameter data and capturing the current frame from the graphics card memory, the system timestamp of the current frame is aligned with the image timestamp acquired by the eye tracker using nearest neighbor interpolation.

[0010] Preferably, the emotional pupil data is obtained using the following method: Before the test begins or during a period of stable light exposure, select a section of length... Based on the data, the parameters of the following linear model are estimated using the least squares method:

[0011] in, Let be the light sensitivity gain coefficient to be solved. The pupil baseline constant, For residual terms; Using the pre-set coefficients By performing decoupling budgeting on signals across all time periods, the pure emotional pupil diameter is obtained. , .

[0012] Preferably, the method for obtaining the emotional pupil data further includes the following steps: Based on input signal The brightness gradient generates an adaptive weighted window; Within the adaptive weighted window, if the decoupled signal still exhibits discontinuous transitions at the window edges, a weighted moving average filter is used to smoothly connect the small neighborhood.

[0013] Preferably, when generating the adaptive weighted window: calculate the input signal The first reciprocal of the derivative; when the absolute value of the derivative exceeds a preset threshold, a correction mechanism is triggered, utilizing the time difference. A time window affected by this process is generated by shifting the time interval forward, and its time interval is defined as follows: ,in, Indicates brightness stimulus signal The moment when a step change occurs or the derivative exceeds the threshold. This indicates the preset duration of pupillary light reflection response decay (usually 3-5 seconds).

[0014] Compared with existing technical solutions, the present invention has the following beneficial effects: 1. Improved the ecological validity of clinical assessment This invention uses algorithms to eliminate lighting artifacts, allowing psychological assessments to move beyond the limitations of a laboratory environment with a gray screen. Patients can be monitored while watching videos with rich color and brightness variations, playing games, or performing complex interactive tasks, resulting in emotional data that more closely reflects real-life situations. 2. Achieved high-precision verification with zero hardware cost. This invention creatively utilizes the computer's internal "system log / video memory data" as the true value of illumination, replacing traditional external photometer hardware. This not only reduces equipment costs but also eliminates the problem of external sensors being affected by ambient diffuse reflection interference by directly reading digital signals, resulting in a higher signal-to-noise ratio. 3. Solved the problem of temporal asynchrony of physiological signals. By introducing a neural conduction latency model, the problem of a 200ms time misalignment between "screen brightening" and "eye shrinking" is solved. Compared to traditional synchronous data removal methods, this approach can retain valid data to the greatest extent and avoid information loss due to accidental data deletion. Attached Figure Description

[0015] Figure 1 This is a system logical architecture diagram; Figure 2 For comparison of the effects, the waveforms are shown from top to bottom as waveform A, waveform B, and waveform C. Waveform A shows a sudden increase in screen brightness, waveform B shows the original pupil data followed by a sharp dip (artifact), and waveform C shows the data after processing by this algorithm, where the dip feature is effectively suppressed while retaining subtle fluctuations (emotional features). Detailed Implementation

[0016] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0017] The present invention discloses a method for blind source separation of non-stationary pupil signals based on a Gamma unit impulse response model, which specifically includes the following steps: Step S1: Construct a dual-modal heterogeneous data synchronization stream based on the system's underlying layer. Frame data is captured directly at the output compositing layer of the graphics rendering pipeline using desktop-level screen capture APIs provided by the operating system (e.g., DXGI Desktop Duplication for Windows or X11 / Wayland Composite interface for Linux). Each frame is converted to grayscale and the weighted average brightness of the central region (e.g., a 10-degree field of view) is calculated as the brightness stimulus signal. This brightness stimulation signal It has a built-in system timestamp and uses nearest-neighbor interpolation to align with the timestamps of images acquired by the eye tracker, eliminating clock drift caused by hardware transmission. Physiological feedback flow (input A): via infrared camera at a sampling rate (e.g., 60Hz) Acquire video of the user's eyes, and use edge detection algorithms to extract the original pupil diameter sequence in real time. .

[0018] Visual stimulus stream (input B): Hooks the display buffer data at the operating system rendering layer to calculate the equivalent luminance stimulus value of the current screen's fovea region (e.g., a 10° viewing angle range at the center of the screen) in real time. This step directly reads the digital signal, requiring no hardware optical sensor. Step S2: Establish a personalized "light-pupil" neural conduction convolution model Using the assumption of linear time-invariant systems, a pulse response model of the retina-midbrain-pupil sphincter is constructed. The Gamma distribution function is used to simulate the hysteresis characteristics of neurotransmitter release and muscle contraction.

[0019] Define the pupil response function to light impulses as follows: This function is a Gamma distribution curve that contains two key physiological parameters: Incubation period ( ): refers to the time difference (usually about 200-250ms) between the retina receiving light and the pupillary muscle beginning to contract, as well as shape parameters. (The peak shape of the response curve is determined, usually by taking an integer value, such as...) =2 or =3).

[0020] Constriction-recovery rate: describes the dynamic process of pupil constriction and rebound.

[0021] Theoretical light reflection component The calculation formula is: ,in, This represents the convolution operation.

[0022] Step S3: Adaptive weighted window generation based on brightness gradient Calculate the luminous stimulus flow The first reciprocal of the derivative. When the absolute value of the derivative exceeds a preset threshold (such as... When this occurs, a correction mechanism is triggered.

[0023] Using the latency parameter in step S2 A time window affected by this process is generated by shifting the time interval forward, and its time interval is defined as follows: ,in, Indicates brightness stimulus signal The moment when a step change occurs or the derivative exceeds the threshold. This indicates the preset duration of pupillary light reflection response decay (usually 3-5 seconds). Within this window, it is determined that pupillary changes are mainly driven by illumination, requiring forced correction.

[0024] Step S4: Signal decoupling and reconstruction based on model inversion Sensitivity calibration: Before the start of the test or during a period of stable illumination, select a segment of length... Based on the data, the parameters of the following linear model are estimated using the least squares method:

[0025] in, Let be the light sensitivity gain coefficient to be solved. The pupil baseline constant, This is the residual term.

[0026] Convolutional subtraction denoising: using predefined coefficients By performing decoupling budgeting on signals across all time periods, the pure emotional pupil diameter is obtained. :

[0027] Residual smoothing (optional): If, within the window marked in step S3, the decoupled signal still exhibits discontinuous jumps at the window edges (indicating an error in the model fitting), then a weighted moving average filter is used to smooth the connection of this small neighborhood.

Claims

1. A method for blind source separation of nonstationary pupil signals based on a Gamma unit impulse response model, characterized in that, Includes the following steps: Before the test began, a completely black screen and a completely white screen were played, and the maximum pupil diameter of the subjects was recorded respectively. and minimum pupil diameter ,like If the ambient light interference is too great, the process is terminated; otherwise, proceed to the next step. Raw pupil diameter data is collected using an eye tracker, and the current frame is captured from the graphics card's memory using an operating system hook function. A 10-degree viewing angle region centered on the fixation point is defined as the effective brightness stimulus area, and the weighted average grayscale value of the pixels within this region is calculated as the input signal. A linear time-invariant system model is constructed, and the Gamma probability density function is used as the system's unit impulse response. , ,in: For shape parameters; The time difference between the retina receiving light and the pupillary muscle beginning to contract is a time constant; the input signal... and Discrete convolution operations are performed to obtain the predicted light reflection components, and the optimal scaling factor is solved using the least squares method. Subtracting the predicted light reflection component from the original pupil data yields the emotional pupil data after artifact removal.

2. The nonstationary pupil signal blind source separation method based on the Gamma unit impulse response model as described in claim 1, characterized in that, The playback time for both the pure black screen and the pure white screen is 5 seconds.

3. The nonstationary pupil signal blind source separation method based on the Gamma unit impulse response model as described in claim 1, characterized in that, When collecting raw pupil diameter data using an eye tracker, the sampling frequency should be no less than 60Hz.

4. The non-stationary pupil signal blind source separation method based on the Gamma unit impulse response model as described in claim 1, characterized in that, When collecting raw pupil diameter data and capturing the current frame from the graphics card memory, the system timestamp of the current frame is aligned with the image timestamp collected by the eye tracker using nearest neighbor interpolation.

5. The non-stationary pupil signal blind source separation method based on the Gamma unit impulse response model as described in claim 1, characterized in that, Specifically, the emotional pupillary data were obtained using the following method: Before the start of the test or during a period of stable light, a segment of length [missing information] was selected. Based on the data, the parameters of the following linear model are estimated using the least squares method: ,in, Let be the light sensitivity gain coefficient to be solved. The pupil baseline constant, For the residual term; using predefined coefficients By performing decoupling budgeting on signals across all time periods, the pure emotional pupil diameter is obtained. , .

6. The nonstationary pupil signal blind source separation method based on the Gamma unit impulse response model as described in claim 5, characterized in that, The method for obtaining the emotional pupil data further includes the following steps: based on the input signal The brightness gradient is used to generate an adaptive weighted window. Within the adaptive weighted window, if the decoupled signal still has discontinuous jumps at the window edge, a weighted moving average filter is used to smoothly connect the small neighborhood.

7. The method for blind source separation of non-stationary pupil signals based on the Gamma unit impulse response model as described in claim 6, characterized in that, When generating an adaptive weighted window: calculate the input signal The first reciprocal of the derivative; when the absolute value of the derivative exceeds a preset threshold, a correction mechanism is triggered, utilizing the time difference. A time window affected by this process is generated by shifting the time interval forward, and its time interval is defined as follows: ,in, Indicates brightness stimulus signal The moment when a step change occurs or the derivative exceeds the threshold. This indicates the preset duration of pupil light reflection response decay.

Images