Advertisement effect evaluation method and system based on eye movement data visualization

Abstract

This invention relates to the field of commercial information processing technology and discloses a method and system for evaluating advertising effectiveness based on eye-tracking data visualization. Addressing the problems of existing evaluation methods' difficulty in objectively quantifying visual attention, and the high cost, complex operation, and lack of customized advertising analysis functions of general-purpose eye-tracking analysis software, this invention configures an eye-tracking device and performs multi-point calibration to simultaneously collect data such as fixation point coordinates, duration, and pupil diameter. Through multidimensional cleaning, fixation event extraction, and coordinate normalization, a fixation heatmap is generated using Gaussian kernel density estimation, and a fixation trajectory map is generated through temporal interpolation. The invention also automatically calculates indicators such as the proportion of fixation duration in the core focal area, overlap, and the proportion of blind spots. This invention achieves full automation from data collection to report generation, objectively and accurately revealing the distribution of visual attention in advertising, transforming physiological data into intuitive design basis, and providing a standardized and repeatable quantitative tool for advertising effectiveness evaluation.

Description

Technical Field

[0001] This invention relates to the field of commercial information processing technology, and in particular to a method and system for evaluating advertising effectiveness based on eye-tracking data visualization. Background Technology

[0002] In the field of digital marketing and advertising design, how to objectively and accurately evaluate the visual appeal of advertising images to the audience has always been a research hotspot and an industry pain point. The visual effect of advertising directly affects consumers' cognition, emotional resonance, and subsequent purchase decisions. Currently, the mainstream advertising effectiveness evaluation methods are mainly divided into two categories: one is statistical methods based on user behavior data, such as click-through rate (CTR) analysis; the other is questionnaire surveys or self-reporting methods based on subjective user feedback.

[0003] However, all of the aforementioned existing technologies have significant limitations. Specifically, while click-through rate (CTR) statistics can reflect macro-level user responses, they cannot reveal specific visual behaviors of users when viewing advertisements, such as which element in the advertisement the user focused on, for how long, and in what order the information was viewed—these are procedural details. Questionnaire surveys and self-report methods rely heavily on users' memory and subjective judgment, and are easily influenced by factors such as recall bias and social expectation effects, making it difficult to objectively and in real-time reflect the audience's true unconscious visual attention behavior.

[0004] The emergence of eye-tracking technology offers a new solution to the aforementioned problems. Eye tracking can record physiological data such as fixation point coordinates, fixation duration, saccade trajectory, and pupil changes when the human eye views visual stimuli with millisecond-level precision, thereby objectively and quantitatively reflecting the distribution of the user's visual attention and cognitive processing. In recent years, some studies have attempted to introduce eye-tracking technology into the field of advertising effectiveness evaluation, identifying visual focal points and blind spots in advertisements by analyzing users' gaze heatmaps or trajectory maps of advertising images.

[0005] However, existing general-purpose eye-tracking analysis software (such as Tobii Pro Lab and SMI BeGaze) is mainly geared towards general research scenarios such as psychology or usability testing, and has the following technical shortcomings:

[0006] 1. High system cost and complex operation: Professional eye-tracking analysis software is usually expensive, and the experimental procedures and parameter configurations are complex, making it difficult for non-professionals to get started quickly, which limits its widespread application in commercial scenarios such as advertising design.

[0007] 2. Lack of customized functions for advertising evaluation scenarios: The heat maps and trajectory maps output by general software are relatively primitive and lack dedicated analytical indicators for evaluating advertising visual attention (such as the proportion of viewing time in the core area of ​​the advertisement, the degree of overlap of focus among multiple subjects, the proportion of visual blind spots, etc.), which have limited guiding role in advertising design optimization.

[0008] 3. Disconnect between data processing and visualization processes: From raw eye-tracking data to the final visualization charts, multiple software programs or multiple manual steps are often required, making it difficult to form a standardized and automated experimental loop, which reduces evaluation efficiency and the reproducibility of results.

[0009] Therefore, there is an urgent need to design an eye-tracking data visualization system for evaluating visual attention in advertising. This system should be characterized by low cost, ease of operation, and customizable analysis. It should be able to automatically complete the entire process from eye-tracking data collection, cleaning, and analysis to generating intuitive gaze heatmaps and trajectory maps, and output key quantitative indicators. This would provide advertising designers with an objective and scientific tool for evaluating visual attention, making up for the shortcomings of existing technologies in the field of advertising effectiveness evaluation. Summary of the Invention

[0010] This invention provides an advertising effectiveness evaluation method and system based on eye-tracking data visualization. The technical problem it solves is that existing advertising effectiveness evaluation methods are difficult to objectively and quantitatively reveal the audience's true visual attention, while general eye-tracking analysis software is costly, complex to operate, and lacks customized analysis functions for advertising scenarios.

[0011] To address the above technical problems, this invention provides a method for evaluating advertising effectiveness based on eye-tracking data visualization, comprising the following steps:

[0012] Configure and calibrate the eye-tracking device and display terminal to determine the pupil-screen coordinate mapping relationship;

[0013] During the process of the subject naturally viewing the advertisement image, the eye-tracking device synchronously collects the gaze point coordinates, gaze duration, and pupil diameter to form an original eye-tracking dataset, wherein the gaze point coordinates are calculated in real time based on the pupil-screen coordinate mapping relationship.

[0014] The raw eye-tracking data is cleaned and then normalized to obtain an effective fixation sequence;

[0015] Based on the effective gaze point sequence, a gaze trajectory map is generated by linear interpolation, and a gaze heatmap is generated by Gaussian kernel density estimation.

[0016] Based on the effective gaze point sequence, the gaze trajectory map, and the gaze heatmap, advertising effectiveness evaluation indicators are calculated, and a visual advertising effectiveness evaluation report is generated.

[0017] Furthermore, configuring and calibrating the eye-tracking device and display terminal to determine the pupil-screen coordinate mapping relationship specifically includes:

[0018] Configure an infrared pupil-corneal reflection eye-tracking device and display terminal, and set up the data acquisition environment;

[0019] Multiple calibration target points are presented sequentially at preset positions on the display terminal, guiding the subject to look at each calibration target point in turn, and judging whether the calibration is passed based on the gaze deviation;

[0020] Based on the relative offset vector between the pupil center and the corneal reflection point, and the pupil diameter, a multinomial regression model is used to establish a mapping function from the pupil-corneal reflection feature space to the screen pixel coordinate space, which serves as the pupil-screen coordinate mapping relationship.

[0021] Furthermore, the simultaneous acquisition of fixation point coordinates, fixation duration, and pupil diameter specifically includes:

[0022] The advertising images to be evaluated are loaded onto the display terminal in a preset order, and then processed for size normalization, color space conversion, and brightness equalization.

[0023] A central gaze point is inserted before each advertisement image is displayed to reset the initial gaze position;

[0024] During the process of subjects naturally viewing the advertisement images, the pixel coordinates of the fixation point with timestamps, the duration of fixation, and the pupil diameter are collected in real time and simultaneously, and data validity labels are attached to form the original eye-tracking dataset.

[0025] Furthermore, the step of cleaning and normalizing the raw eye-tracking data to obtain an effective fixation sequence specifically includes:

[0026] Detect and remove blink artifacts, large head movement interference data, and abnormal data with low confidence, coordinates outside the screen range, or pupil diameter outside the physiological range;

[0027] The remaining data is clustered into gaze events by using a velocity threshold method, and gaze events with a duration less than a preset threshold are filtered out to obtain a coarse gaze point sequence.

[0028] The pixel coordinates in the coarse fixation point sequence are normalized to a unit interval according to the screen width and height to obtain the effective fixation point sequence.

[0029] Furthermore, the detection and removal of blink artifact data specifically includes: jointly determining blink events based on the rate of change of pupil area between adjacent frames and the abrupt change in the position of corneal reflex points, and removing all eye movement data within a preset time window before and after the blink event.

[0030] Furthermore, the detection and removal of large head movement interference data specifically includes: acquiring head displacement velocity and rotation angular velocity; when the displacement velocity or rotation angular velocity exceeds the corresponding threshold, it is determined as a head interference event, and all eye movement data during the duration of the interference event and within a preset time before and after it is removed.

[0031] Furthermore, the generation of the gaze trajectory map through linear interpolation specifically includes:

[0032] The gaze events are extracted from the effective gaze sequence and sorted in ascending order by timestamp to obtain an ordered gaze sequence.

[0033] The spatial location of each gaze event is used as a trajectory node;

[0034] Multiple intermediate points are generated by uniformly interpolating over time between adjacent trajectory nodes. The points are connected to form an interpolation path. The trajectory nodes and the interpolation path are then overlaid on the original advertisement image to form a gaze trajectory map.

[0035] Furthermore, the generation of the gaze heatmap through Gaussian kernel density estimation specifically includes:

[0036] Using the normalized coordinates of each fixation event in the effective fixation point sequence as the center and the fixation duration as the weight, the attention density value of each point on the normalized plane is calculated by a two-dimensional Gaussian kernel function.

[0037] The normalized plane is discretized into a grid, and the density value of each grid cell is calculated to obtain a two-dimensional density matrix;

[0038] After linearly normalizing the original density values ​​in the two-dimensional density matrix, they are converted into color values ​​according to a preset color mapping table to generate a heat map layer, which is then superimposed on the original advertising image in a semi-transparent manner to form a gaze heat map.

[0039] Furthermore, the calculation of advertising effectiveness evaluation indicators and the generation of a visualized advertising effectiveness evaluation report specifically includes:

[0040] Aggregate the gaze heatmaps of all subjects, and automatically identify the region with the highest density as the core focus region through threshold segmentation and connected component analysis;

[0041] Calculate the percentage of total fixation time in the core focus area for all subjects out of the total fixation time for the entire advertisement, as the core focus area fixation time percentage;

[0042] The percentage of subjects whose gaze fell into the core focal area at least once was calculated as the focal area overlap.

[0043] The area with a density value lower than the blind zone density threshold in the aggregated heat map is defined as the advertising blind zone. The proportion of pixels of this advertising blind zone to the entire advertising image is calculated as the visual blind zone proportion.

[0044] A visual advertising effectiveness evaluation report is generated based on the above indicators.

[0045] This invention also provides an advertising effectiveness evaluation system based on eye-tracking data visualization, the key of which includes:

[0046] The experimental configuration and calibration module is used to configure the eye-tracking device and display terminal and perform calibration to establish the pupil-screen coordinate mapping relationship.

[0047] The data acquisition module is used to simultaneously collect fixation point coordinates, fixation duration, and pupil diameter while subjects view advertising images, forming a raw eye-tracking dataset;

[0048] The data cleaning and gaze detection module is used to clean the raw eye movement data, extract gaze events and normalize coordinates, and output a valid gaze point sequence.

[0049] The heatmap and trajectory map generation module is used to generate gaze trajectory maps based on effective gaze point sequences through linear interpolation and to generate gaze heatmaps through Gaussian kernel density estimation.

[0050] The indicator calculation and reporting module is used to calculate advertising effectiveness evaluation indicators based on effective gaze point sequences, gaze trajectory maps, and gaze heatmaps, and generate visual evaluation reports.

[0051] This invention provides a method and system for evaluating advertising effectiveness based on eye-tracking data visualization. By configuring eye-tracking devices and display terminals and performing multi-point calibration, a precise pupil-screen coordinate mapping relationship is established, ensuring data acquisition accuracy. Based on this, multi-dimensional eye-tracking parameters such as fixation point coordinates, fixation duration, and pupil diameter are simultaneously collected when subjects view advertising images, comprehensively recording real visual behavior. The raw eye-tracking data is then subjected to multi-dimensional cleaning to remove blink artifacts, large head movements, and low-confidence abnormal data. A velocity thresholding method is used to extract valid fixation events, and coordinates are normalized to a unified space, effectively reducing environmental interference and physiological noise. Subsequently, a fixation heatmap is generated using two-dimensional Gaussian kernel density estimation with fixation duration weighting, intuitively presenting the distribution of attention intensity. Simultaneously, a fixation trajectory map is generated through temporal interpolation, clearly showing the browsing order and path. Finally, quantitative indicators such as the fixation duration percentage of the core focal area, the overlap of focal areas, and the percentage of visual blind spots are automatically calculated, and a visual evaluation report is output. This method achieves a fully automated closed loop from data collection, cleaning, analysis to report generation. It can objectively and accurately reveal the distribution patterns and potential problems of visual attention in advertising, transforming abstract physiological data into intuitive design basis, and providing a standardized and repeatable quantitative tool for advertising effectiveness evaluation. Attached Figure Description

[0052] Figure 1 A flowchart illustrating an advertising effectiveness evaluation method based on eye-tracking data visualization provided in an embodiment of the present invention;

[0053] Figure 2 A module architecture diagram of an advertising effectiveness evaluation system based on eye-tracking data visualization provided in an embodiment of the present invention;

[0054] Figure 3 This is a schematic diagram comparing the gaze trajectory diagrams and gaze heatmaps of three subjects for four advertising images in an embodiment of the present invention;

[0055] Figure 4 This is a bar chart showing the percentage of average gaze duration in the core focus area of ​​four advertising images in this embodiment of the invention.

[0056] Figure 5 This is a bar chart showing the overlap of the focal areas of four advertising images in an embodiment of the present invention.

[0057] Figure 6 This is a pie chart showing the percentage of visual blind spots in four advertising images from an embodiment of the present invention. Figure 6 (a) Figure 6 (b) Figure 6 (c) Figure 6 The middle (d) is a pie chart showing the percentage of visual blind spots for P0, P1, P2, and P3, respectively. Detailed Implementation

[0058] The embodiments of the present invention are described in detail below with reference to the accompanying drawings. The embodiments are given for illustrative purposes only and should not be construed as limiting the present invention. The accompanying drawings are for reference and illustration only and do not constitute a limitation on the scope of patent protection of the present invention, because many changes can be made to the present invention without departing from the spirit and scope of the present invention.

[0059] The advertising effectiveness evaluation method based on eye-tracking data visualization provided in this invention embodiment, such as... Figure 1 The flowchart shown includes the following steps:

[0060] S1. Environment Configuration and Eye Tracking Calibration: Configure the eye tracking device and display terminal and perform calibration to determine the pupil-screen coordinate mapping relationship;

[0061] S2. Synchronous eye-tracking data acquisition: During the subject's natural viewing of the advertisement image, the eye-tracking device synchronously collects the gaze point coordinates, gaze duration, and pupil diameter to form the original eye-tracking dataset. The gaze point coordinates are calculated in real time based on the pupil-screen coordinate mapping relationship.

[0062] S3. Eye movement data cleaning and normalization: The raw eye movement data is cleaned and then normalized to obtain an effective fixation sequence.

[0063] S4. Heatmap and trajectory map generation: Based on the time-series gaze point sequence, gaze trajectory maps are generated by linear interpolation, and gaze heatmaps are generated by Gaussian kernel density estimation.

[0064] S5. Calculation and Output of Advertising Attention Evaluation Indicators: Based on the effective gaze point sequence, gaze trajectory map and gaze heatmap, calculate advertising effectiveness evaluation indicators and generate a visualized advertising effectiveness evaluation report.

[0065] This invention provides an advertising effectiveness evaluation method based on eye-tracking data visualization. First, the eye-tracking device undergoes professional configuration and multi-point calibration to effectively ensure the accuracy and consistency of subsequent data collection. Based on this, multi-dimensional eye-tracking parameters such as fixation point coordinates, fixation duration, and pupil diameter are simultaneously collected when subjects view advertising images, thus comprehensively recording their actual visual behavior. Subsequently, the raw eye-tracking data is cleaned and normalized to significantly reduce environmental interference and physiological noise, ensuring the quality and standardization of the input data. Then, Gaussian kernel density estimation and trajectory interpolation algorithms are used to generate intuitive fixation heatmaps and fixation trajectory maps, transforming discrete fixation points into high-resolution probability density fields and spatiotemporal paths, clearly presenting the audience's attention intensity and browsing order for different areas of the advertisement. Finally, quantitative indicators such as the fixation percentage of the core focus area, the overlap of focus among multiple subjects, and the proportion of visual blind spots are automatically calculated, and a visual evaluation report is output. The entire method achieves a closed-loop process from data collection, cleaning, and modeling to indicator output, objectively and accurately revealing the distribution patterns and potential problems of visual attention in advertising, providing repeatable and quantifiable physiological data for advertising design optimization.

[0066] (1) Step S1: Environment setup and eye-tracking calibration

[0067] This step specifically includes:

[0068] S11. Configure the hardware environment required for eye-tracking data acquisition, including: configuring the eye-tracking device, display terminal, and control acquisition environment.

[0069] The eye-tracking device uses an infrared pupil-corneal reflection eye tracker (such as Tobii Pro Nano, 7invensun A8, etc.), with a sampling rate of no less than 60 Hz and a spatial resolution of ≤0.5°, to capture the pupil center coordinates and corneal reflection point position of the subject in real time.

[0070] The display terminal uses a high-resolution (no less than 1920×1080 pixels) LCD screen, and the screen size and viewing distance are set according to standardized requirements (usually 60~80 cm from the subject's eyes, and the horizontal viewing angle of the screen ≤30°) to ensure that the eye-tracking device can stably capture pupil and corneal reflection characteristics.

[0071] The data acquisition environment is set in an indoor space with uniform lighting and no strong reflections or background interference. Direct sunlight or high-frequency light should be avoided to reduce infrared reflection noise.

[0072] S12. Before formally collecting eye-tracking data from the advertising images, a standardized calibration process is performed to establish a precise mapping between the subject's gaze point coordinates and the screen pixel coordinates. The specific steps are as follows:

[0073] Multi-point calibration: Calibration target points (such as red circles or cartoon icons) are sequentially displayed at preset positions (usually 5, 9 or 13 points) within the screen display area, with each target point displayed for no less than 1 second;

[0074] Gaze guidance: Subjects are instructed to gaze at each calibration target point in sequence through voice or text prompts, and to keep their heads stable during gaze maintenance;

[0075] Error calculation: The eye-tracking device records the pupil-corneal reflection vector corresponding to each calibration point in real time, compares it with the screen coordinates of the preset target point, and calculates the average fixation deviation;

[0076] Acceptance / Retry Judgment: If the average gaze deviation is ≤0.5° and the maximum deviation of all calibration points is ≤1.0°, the calibration is considered successful; otherwise, recalibration is prompted until the accuracy requirements are met.

[0077] S13: Generate a pupil-screen coordinate mapping relationship based on the relative offset vector between the pupil center and the corneal reflection point and the pixel coordinates of the fixation point on the screen.

[0078] After calibration, the system automatically generates a mapping function from the pupil-corneal reflectance feature space to the screen pixel coordinate space. This mapping typically uses a polynomial regression model (such as a second-order polynomial), formally represented as:

[0079] ,

[0080] in, The pixel coordinates of the gaze point on the screen. This is the relative offset vector between the pupil center and the corneal reflection point (in the built-in image coordinate system of the eye-tracking device). The diameter of the pupil. , is a polynomial function obtained by fitting the calibration sample points.

[0081] Through the above-mentioned equipment configuration, precise calibration and mapping modeling, it is possible to ensure that the eye movement data collected subsequently has high precision (deviation ≤0.5°) and high consistency, eliminating the influence of individual differences (such as differences in eye geometry parameters and head position of different subjects) on the calculation of fixation point coordinates, and providing a solid pre-condition for the collection of eye movement data during the viewing of advertising images in step S2.

[0082] (2) Step S2: Synchronous acquisition of eye-tracking data

[0083] This step specifically includes:

[0084] S21. Control the display terminal to display advertising images and initialize the eye-tracking device.

[0085] After completing the S1 environment configuration and eye-tracking calibration, the formal data acquisition phase begins. The advertising images (stimuli) to be evaluated are loaded onto the display terminal in a preset order. The resolution of each advertising image is uniformly adjusted to the same screen resolution as in the calibration phase (e.g., 1920×1080 pixels), and standardized preprocessing is performed before display, including size normalization, color space conversion, and brightness equalization, to eliminate potential interference from display differences between different advertising images on the eye-tracking data.

[0086] The presentation time of each advertisement image is preset to a fixed duration (usually 5 to 10 seconds). This duration should be sufficient for the subject to complete a natural visual browsing cycle, while avoiding visual fatigue or distraction caused by excessive presentation time.

[0087] In addition, a central fixation point lasting 500-1000 milliseconds is inserted before each advertisement image is displayed to reset the subject's initial fixation position to zero, ensuring that the fixation starting point is consistent for each image, which facilitates subsequent trajectory analysis.

[0088] S22. The eye-tracking device collects multidimensional eye-tracking data in real time, including timestamps, fixation coordinates (the subject's two-dimensional fixation position on the screen, pixel coordinates), fixation duration (the duration of fixation at each fixation point), and pupil diameter.

[0089] Multidimensional eye movement data were simultaneously collected by an eye-tracking device while the subjects naturally viewed the advertisement images.

[0090] The timestamp refers to the absolute time (milliseconds) of each frame of data acquisition, synchronized with the system clock. The sampling interval is determined by the eye tracker's sampling rate (e.g., 60 Hz corresponds to approximately 16.67 ms / frame).

[0091] Pupil diameter refers to the diameter of the pupils of both eyes (in millimeters or pixels), which is extracted in real time by the infrared imaging and image processing algorithm built into the eye-tracking device, reflecting the subject's physiological arousal or cognitive load.

[0092] S23. Construct the original eye-tracking dataset based on the collected multidimensional eye-tracking data.

[0093] During the acquisition process, the system writes eye-tracking data for each frame to local storage or memory buffer in real time. Each data record contains at least the following fields: global timestamp (milliseconds) of the sampling time, unique identifier of the currently presented advertisement image, X-pixel coordinates of the gaze point, Y-pixel coordinates of the gaze point, diameter of the left pupil (mm), diameter of the right pupil (mm), and data validity indicator (0=invalid, 1=valid).

[0094] After the subjects have finished viewing all the advertising images, the system exports all the raw eye-tracking data records into a standard format file (such as CSV, Parquet, or a dedicated binary format), forming a raw eye-tracking dataset that can be used for subsequent processing.

[0095] (3) Step S3: Eye-tracking data cleaning and normalization

[0096] Step S3 performs multi-dimensional cleaning and coordinate normalization on the raw eye-tracking dataset acquired in step S2, removing invalid or interfering data, and unifying the gaze point coordinates to a standardized feature space, outputting a valid gaze point sequence that can be used for subsequent visualization and quantitative analysis. Step S3 specifically includes:

[0097] S31. Detect and remove blink artifacts from the original eye-tracking dataset.

[0098] Blinking causes a sudden decrease in pupil area and a temporary disappearance of the corneal reflex point, producing abnormal eye movement signal fragments. This step uses a combination of pupil area change rate and corneal reflex signal to detect blinking events, specifically including:

[0099] The pupil area (converted from the pupil diameter) is calculated for each frame, and the rate of change of area between adjacent frames is also calculated. When the pupil area suddenly drops to less than 30% of the normal value within 3-5 consecutive frames, and at the same time the distance of the corneal reflector point changes abruptly (e.g., the coordinate offset of the corneal reflector point exceeds 5% of the screen width), it is determined to be a blink event. All eye movement data within a 150 ms time window before and after the blink event are removed to eliminate signal distortion caused by blinking and unreliable data during the recovery period before and after it.

[0100] S32. Filter out large head movement interference data from the remaining data in the original eye-tracking dataset.

[0101] Significant head movements by the subject can cause the pupil-corneal reflex vector to exceed the linear operating range of the eye-tracking device, resulting in systematic errors. This step identifies interference by measuring the head displacement velocity and the rate of change of rotation angle, specifically including:

[0102] The three-dimensional displacement velocity of the head is obtained from the head posture sensor (or by the algorithm built into the eye tracker). and rotational angular velocity .when >50mm / s or An eye movement rate >15° / s is considered a head-interference event. All eye movement data during the duration of this interference event and within 100 ms before and after it are discarded.

[0103] S33. Filter out abnormal data in the remaining data of the original eye-tracking dataset.

[0104] Based on the confidence score, spatial coordinate rationality, and physiological parameter range provided by the eye-tracking device, low-quality data was further eliminated:

[0105] Frames with a validity score below 0.8 were removed. The validity score was given by the eye-tracking device based on indicators such as pupil detection quality and corneal reflection clarity.

[0106] Remove data whose gaze point coordinates are outside the screen display range or whose coordinates are infinity or non-numerical;

[0107] Exclude abnormal values ​​with a pupil diameter less than 1.0 mm or greater than 8.0 mm (these values ​​exceed the normal physiological range of human pupils and are usually caused by detection errors);

[0108] Calculate the instantaneous angular velocity between adjacent fixation points. If it exceeds 800° / s (exceeding the human physiological limit), it is judged as a data jump, and the corresponding jump point is removed.

[0109] S34. Extract the coarse fixation sequence from the remaining data of the original eye-tracking dataset.

[0110] After the cleaning described in steps S31 to S33, a discrete sequence of original sampling points is obtained. These sampling points need to be categorized into fixation events and saccade events to extract meaningful fixation point sequences. This step uses the velocity thresholding method (I-VT), specifically:

[0111] Calculate the angular velocity of each sampling point (equal to the viewing angle converted from the Euclidean distance between adjacent sampling points) and divide it by the time interval;

[0112] Set a speed threshold (a common value is 30° / s, which can be adjusted according to the actual equipment and subjects).

[0113] When the angular velocities of multiple consecutive sampling points are all less than the velocity threshold, these sampling points are classified into a single gaze event.

[0114] For each gaze event, calculate the center position of its gaze point (which can be the median or weighted average of the coordinates of all sampled points). The gaze duration is the time difference between the first and last frames in the event, and the coordinates of the gaze point are the spatial center of the event.

[0115] Filter out fixation events with a duration of less than 50 ms (which are usually considered as microsaccades or noise).

[0116] After the above classification, the original sampling points are compressed into a series of gaze events, each of which contains a timestamp (start time or intermediate time), spatial coordinates and gaze duration, forming a coarse gaze point sequence.

[0117] S35. Normalize the coordinates of the coarse fixation point sequence to obtain the effective fixation point sequence.

[0118] Different advertising images may have different resolutions, and the original gaze coordinates are absolute pixel values, which are not convenient for aggregated analysis across images and subjects. This step normalizes the gaze coordinates to a uniform unit space [0,1]×[0,1]:

[0119] Let the screen width be pixels, height is Pixels, the raw coordinates for each frame or each gaze event. Perform normalization transformation: ;

[0120] If the origin of the original coordinate system output by the eye-tracking device is in the lower left corner, while the origin of the screen rendering coordinate system is in the upper left corner, then an additional vertical axis flip is required: .

[0121] Normalized coordinates All data are within the range [0,1][0,1], allowing advertising images of different resolutions and eye movement data from different subjects to be compared and superimposed in the same standard space.

[0122] After the above cleaning, classification, and normalization, the final output is a valid fixation sequence. Each record contains the following fields: unique identifier of the advertisement image, unique identifier of the subject, fixation number under the advertisement image (in ascending order of time), global timestamp of the start or intermediate moment of fixation, normalized fixation x-coordinate (0~1), normalized fixation y-coordinate (0~1), and fixation duration (milliseconds). This sequence can be used as direct input for generating heatmaps and trajectory maps in S4, and can also be used for calculating various quantitative indicators in S5.

[0123] Step S3, through the aforementioned multidimensional cleaning and normalization processes, systematically removes blink artifacts, head movement interference, and low-confidence outliers from the original eye-tracking data, transforming discrete sampling points into semantically meaningful gaze event sequences while simultaneously achieving unified standardization of the coordinate space. The processed effective gaze point sequences exhibit low noise, uniform format, and clear visual behavioral semantics, providing a reliable data foundation for subsequent high-precision heatmap generation, trajectory plotting, and attention quantification analysis.

[0124] (4) Step S4: Generation of heat map and trajectory map

[0125] Step S4 uses the effective gaze point sequence output from step S3 as input, employing a temporal interpolation algorithm to generate a gaze trajectory map reflecting the visual browsing order, and simultaneously using a two-dimensional Gaussian kernel density estimation algorithm to generate a gaze heatmap reflecting the spatial distribution of visual attention. These two graphs reveal the visual attention characteristics of the audience towards the advertising content from different dimensions.

[0126] Fixation trajectory maps are used to visually demonstrate the movement and sequence of a subject's gaze over time while viewing an advertisement image, thereby revealing visual search strategies and information browsing paths. The steps to generate a fixation trajectory map include:

[0127] S41A. Extract the set of fixation times from the effective fixation point sequence and sort them in ascending order according to the timestamp to obtain an ordered fixation point sequence.

[0128] Extract the set of gaze events corresponding to the current advertisement image from the effective gaze point sequence. ,in The total number of gaze events on this advertisement image. Each gaze event... Includes normalized coordinates timestamp (The start or middle moment of the fixation event) and fixation duration Will be based on timestamps Arrange in ascending order to obtain an ordered fixation point sequence.

[0129] S42A, Use the spatial location of each gaze event as a trajectory point.

[0130] Trajectory diagrams typically use the spatial location of the observed event as the key node.

[0131] To visually reflect the differences in fixation duration, each node can be further differentiated and encoded. The encoded content includes:

[0132] Node position: Take As coordinates of trajectory nodes;

[0133] Node radius : with fixation duration Positive correlation (e.g.) The longer the fixation time, the larger the nodal circle becomes. As the reference radius, This is an adaptive adjustment coefficient, which is positively correlated with the size of the entire advertisement image; the larger the advertisement image, the better. The larger.

[0134] Node colors: Gradual coloring in chronological order (e.g., light green → yellow → red to represent time from morning to night).

[0135] S43A: Perform linear interpolation between gaze points to generate a smooth motion path.

[0136] To simulate the continuous movement of the visual focus between adjacent trajectory nodes and avoid the visual discontinuity caused by straight jumps between nodes, this process introduces linear interpolation between adjacent trajectory nodes to generate a smooth movement path.

[0137] Set adjacent fixation points and The coordinates are respectively and The time interval is Generate by uniform interpolation over time between the two. The interpolation function for the intermediate points is:

[0138] ,

[0139] in , This represents the time corresponding to the interpolation point. During actual rendering, the coordinates of an interpolation point are calculated at fixed time steps (e.g., 10-20 ms), and these points are connected sequentially to form the interpolation path. Finally, the interpolation path is rendered as a curve with arrows (the arrow direction indicates the direction of view movement), or directly drawn as a continuous polyline.

[0140] S44A. Overlay the above trajectory nodes and interpolation paths onto the original advertisement image to form the final gaze trajectory map.

[0141] To improve readability, you can optionally add:

[0142] Sequence label: Mark each node with a sequential number (1, 2, 3, …) to clearly indicate the order of attention;

[0143] Transparent background: The trajectory lines are semi-transparent to avoid obscuring the advertising content.

[0144] A gaze heatmap is used to reflect the intensity of visual attention received at different locations on an advertising image. Areas with denser gazes and longer gaze durations have warmer heatmap colors (red, yellow), while areas with lower attention levels have cooler colors (blue, purple) or are transparent. The steps to generate a gaze heatmap include:

[0145] S41B Calculate the attention density value for each gaze event in the effective gaze sequence.

[0146] Suppose that the effective fixation sequence has Each gaze event The normalized coordinates are The duration of fixation is (Unit: milliseconds). Any point on the normalized plane [0,1]×[0,1]. Attention density value at Defined as the sum of the probability densities contributed by each gaze event, it is estimated using a weighted two-dimensional Gaussian kernel density formula:

[0147] ,

[0148] in, For the first The weights of each gaze event are typically chosen based on the gaze duration. As a weight, the longer the gaze time, the greater its contribution to the surrounding density field; It is a two-dimensional Gaussian kernel function (standard normal distribution); , This is a bandwidth parameter that controls the smoothness of the heatmap.

[0149] Using an isotropic Gaussian kernel ( ), kernel function The specific form is as follows:

[0150] .

[0151] bandwidth The value of directly affects the resolution and smoothness of the heatmap. Generally, it is set to . (Relative to the normalized unit length), adaptive estimation can also be performed using the Silverman rule:

[0152] ,

[0153] in, denoted as the standard deviation of the gaze point coordinates.

[0154] S42B discretizes the normalized plane [0,1]×[0,1] into a high-resolution grid.

[0155] To facilitate rendering and display, the normalized plane [0,1]×[0,1] is discretized into a high-resolution mesh, and the mesh size is usually set to [0,1]×[0,1]. For example, a 1024×1024 grid can be used (to match the resolution of the original advertisement image). For the center point of each grid cell, its density value is calculated according to the formula above, resulting in a two-dimensional density matrix. .

[0156] To improve computational efficiency, a fast Gaussian transform or truncating only at each gaze point can be used. The influence area is calculated, ignoring the minor contribution of distant gaze points to the current grid.

[0157] S43B, the density matrix The original density values ​​are linearly normalized to the range [0,1], and the normalized density values ​​are converted to RGB color values ​​using a predefined color mapping table.

[0158] The color mapping table used in this embodiment is shown in Table 1 below.

[0159] Table 1 Color Mapping Table

[0160]

[0161] In practice, predefined mapping relationships (such as jet, hot, or custom thermal color scales in Matplotlib) can be used to linearly interpolate density values ​​to RGB triples.

[0162] S44B. The generated color heatmap layer is overlaid on the original advertisement image in a semi-transparent manner (e.g., transparency 0.6) to form the final visual gaze heatmap.

[0163] Heat maps preserve the original content of the advertisement while highlighting the hot spots that subjects are most interested in.

[0164] Step S4, through the two parallel processes described above, enables the system to automatically generate two complementary visualization charts from the same valid gaze point sequence:

[0165] Gaze trajectory map: In the form of time sequence and spatial path, it intuitively presents the subject's browsing order, jump path and key dwell area of ​​advertising elements, which helps to analyze the rationality of visual search and interference points;

[0166] Attention heatmap: Displays the distribution of attention intensity using color density gradients, quickly locating visual interest centers (hot spots) and completely ignored areas (blind spots) in advertisements.

[0167] (5) Step S5: Calculation and output of advertising attention evaluation index.

[0168] Step S5, based on the effective gaze point sequence output in Step S3, the gaze trajectory map (including the spatial location and sequence of gaze events) generated in Step S4, and the gaze heatmap (two-dimensional density field), automatically calculates three core evaluation indicators and outputs structured evaluation results to quantify the visual appeal and layout rationality of the advertisement. Step S5 specifically includes:

[0169] S51. Based on the gaze heatmaps of all subjects, the region with the highest density is automatically identified as the core focus region through threshold segmentation and connected component analysis.

[0170] The core focus area refers to the key element area in an advertisement image that is expected to attract the audience's attention most, such as the main product, brand logo, call-to-action button, or the advertiser's designated core selling point. First, the gaze heatmaps of all subjects are aggregated (either by averaging or weighted fusion) to obtain an aggregated heatmap.

[0171] Then set the segmentation density threshold to... , This represents the maximum density value from the polymerization thermogram. Extract all values ​​with a density greater than [value missing]. For each pixel, connected components are labeled. The connected component with the largest area is selected as the core focal region, and its minimum bounding rectangle is calculated as the region boundary.

[0172] S52. Calculate the proportion of fixation time in the core focus area based on the effective fixation point sequence and fixation trajectory diagram.

[0173] This metric measures the ability of the core focus area to attract and maintain audience attention, defined as the percentage of total gaze time spent in the core focus area by all subjects out of the total gaze time of the entire advertisement.

[0174] S53. Calculate the overlap of focal regions based on the effective fixation point sequence and fixation trajectory diagram of all subjects.

[0175] This indicator measures the prevalence and consistency of attention to the core focal area among different subjects, and is defined as the percentage of subjects whose gaze falls on the core focal area at least once out of the total number of subjects.

[0176] S54. Generate the percentage of blind spots in advertising based on aggregated heatmaps.

[0177] Specifically, areas with density values ​​below the blind zone density threshold are defined as advertising blind zones, and the blind zone percentage is obtained by calculating the proportion of pixels in the entire advertising image to the number of pixels in the blind zone.

[0178] S55. Generate evaluation index reports for each advertising image in the required formats, such as text, visual charts (bar charts, line charts, and composite pie charts for quantitative analysis, etc.).

[0179] Through the complete calculation process described above, the system automatically extracts three key quantitative indicators from the gaze trajectory map and heatmap: the percentage of gaze time in the core focus area (reflecting the intensity of attraction), the overlap of focus areas (reflecting group consistency), and the percentage of visual blind spots (reflecting layout efficiency). These three indicators complement each other, providing a simple, objective, and comparable basis for optimizing advertising design. The entire process achieves a fully automated mapping from eye-tracking data to advertising effectiveness scores, requiring no subjective human intervention.

[0180] (6) System Design

[0181] This embodiment also provides an advertising effectiveness evaluation system based on eye-tracking data visualization, employing methods such as... Figure 2 The modular architecture shown consists of five core modules, corresponding to the method steps S1 to S5 respectively.

[0182] The experimental configuration and calibration module is responsible for S1: automatically detecting the eye tracker and performing 9-point calibration, calculating the average fixation deviation (threshold 0.5°), establishing a second-order polynomial mapping from the pupil-corneal reflection vector to the screen coordinates, and saving the mapping parameters to the configuration file.

[0183] The data acquisition module corresponding to S2 adopts multi-threaded asynchronous acquisition, synchronously recording the fixation point coordinates, pupil diameter and confidence level of the subject when viewing the advertisement image at a sampling rate of 60 Hz, and realizing millisecond-level time alignment between eye movement data and image switching events through hardware synchronization pulses.

[0184] The data cleaning and gaze detection module implements S3: First, it removes blink artifacts (pupil area drops sharply by 30% and lasts for 3~5 frames), large head movements (velocity > 50 mm / s or angular velocity > 15° / s), and abnormal frames with confidence < 0.8; then, it uses the velocity threshold method (30° / s) to cluster the effective sampling points into gaze events, filters out noise with a duration < 50 ms, and finally normalizes the pixel coordinates to the [0,1]×[0,1] space.

[0185] The heatmap and trajectory map generation module corresponds to S4: Based on the gaze event sequence, a density field is generated by weighted (gaze duration) two-dimensional Gaussian kernel density estimation (bandwidth h=0.06), mapped to a red and yellow heatmap layer and semi-transparently overlaid on the original advertisement image; at the same time, gaze nodes are drawn in chronological order (radius proportional to duration) and smooth trajectory lines with arrows are generated by linear interpolation between adjacent nodes, and output in PNG and SVG formats respectively.

[0186] The indicator calculation and reporting module completes S5: It automatically extracts the core focal region from the group heatmap (threshold segmentation to take the largest connected component), calculates three indicators: the core region fixation time percentage (total fixation time in the region / total fixation time of the whole image), focal region overlap (the proportion of subjects who fixate on the core region at least once), and visual blind spot percentage (the proportion of pixels with a density lower than 5% of the maximum density). Finally, it generates an HTML report containing indicator cards, visualization charts and optimization suggestions through a template engine, and supports one-click export to PDF.

[0187] The modules are connected through standardized data interfaces, enabling fully automated processing from raw eye-tracking data to advertising evaluation reports.

[0188] (7) Experimental Design and Conclusion Analysis

[0189] To verify the effectiveness of the method and system proposed in this embodiment in assessing visual attention in advertising, a controlled experiment was designed and implemented. The experiment employed a within-subjects design, with all participants viewing a set of mobile phone advertising images under identical environmental conditions. Eye-tracking data was simultaneously collected, and a visual analysis report was generated.

[0190] 1) Subject Information

[0191] Thirty university students and working professionals were recruited as participants, including 14 males and 16 females, aged 22-35 years (mean age 26.4 years, standard deviation 3.2 years). All participants had normal visual acuity (both uncorrected and corrected visual acuity ≥0.8), no color blindness or color weakness, and had not previously been exposed to the advertising materials used in this experiment.

[0192] 2) Experimental equipment and environment

[0193] The experimental equipment is configured as follows:

[0194] Eye-tracking device: 7invensun A8 infrared pupil-corneal reflection eye tracker, sampling rate 60 Hz, spatial resolution ≤0.5°, connected to the control host via USB interface;

[0195] Display terminal: 16-inch LCD monitor with a resolution of 2560×1440 pixels and a refresh rate of 60 Hz;

[0196] Control host: Intel Core i7 processor, 16 GB memory, running Windows 10 operating system;

[0197] Software environment: The eye-tracking data visualization system developed in this paper includes an experimental control module, a data acquisition module, and a visualization analysis module.

[0198] The experiment was conducted in an indoor laboratory with uniform lighting and no direct sunlight. Subjects were seated comfortably, with their eyes approximately 50 cm from the center of the monitor screen and a downward viewing angle of approximately 15°. Unnecessary light sources were turned off during the experiment to avoid infrared interference.

[0199] 3) Advertising materials

[0200] Four mobile phone print advertisement images (P0, P1, P2, and P3) were selected as the advertising material. All images had a uniform resolution of 1200×800 pixels and included typical elements such as the product itself, brand logo, promotional text, and background decoration. Each image underwent brightness equalization and size standardization before the experiment to ensure consistent display. The experimental procedure presented the four images to each participant in a random order.

[0201] 4) Experimental Procedure

[0202] The experiment was divided into four stages.

[0203] Phase 1: Equipment Start-up and Calibration

[0204] The eye tracker and control software are activated, and a 9-point calibration procedure is executed to calculate the average fixation deviation. If the deviation is greater than 0.5°, recalibration is performed until the average deviation of all subjects is ≤0.5° and the maximum deviation of a single point is ≤1.0°, to ensure the accuracy of eye movement data acquisition and establish the pupil-screen coordinate mapping relationship.

[0205] Phase Two: Baseline Acquisition

[0206] Before the formal experiment, subjects were shown a blank gray screen (brightness 50 cd / m²) for 5 seconds, and their pupil diameter during this period was recorded as a baseline to obtain an individual baseline for pupil diameter, which was used for subsequent emotional / cognitive load correction.

[0207] Phase 3: Stimulus Material Display and Eye-Tracking Data Acquisition

[0208] Each advertisement image is displayed in full screen for 8 seconds, with a central fixation point (lasting 1 second) inserted between images to correct gaze. Participants are instructed to "view this naturally, as you would when browsing a mobile phone advertisement." The system records fixation point coordinates, fixation duration, pupil diameter, and other data in real time, sampling throughout the entire viewing process. Raw eye movement data of the participants in a natural state is obtained.

[0209] Phase Four: Data Visualization and Metric Calculation

[0210] After the experiment, the system automatically cleans the raw eye-tracking data (removing blink artifacts, head movement interference, and low-confidence data), normalizes it, classifies fixation points, generates heatmaps / trajectory maps, and calculates three core metrics (the percentage of fixation time in the core focal area, the overlap of focal areas, and the percentage of visual blind spots). It outputs a visualized advertising attention assessment report for quantitative analysis.

[0211] Each participant took approximately 8–10 minutes to complete the entire process. A total of 30 valid eye movement data points were collected in the experiment (no participants were excluded due to low data quality).

[0212] 5) Conclusion Analysis

[0213] The collected raw eye-tracking data is processed according to the above method to obtain the group gaze heatmap and gaze trajectory map for each advertisement, and the following three core indicators are calculated: the proportion of gaze duration in the core focus area, the overlap of focus areas, and the proportion of visual blind spots.

[0214] Figure 3The gaze trajectories and gaze heatmaps of three subjects, W, L, and J, for four advertisements are shown (not superimposed on P0 to P3). From Figure 3 It can be seen that although the advertising focus of different subjects is not completely consistent, there is a certain overlap, which can help identify the core focus area.

[0215] Figure 4 The image shows the average view duration percentage of the core focus area in the four advertisement images. From Figure 4 It can be seen that P0 to P3 account for the highest proportions in foldable phone form factor, screen flower pattern, camera module and dual-lamp camera module, respectively, at 42%, 48%, 39% and 45%, with an average of 43.5%, indicating that this method can accurately capture the most attractive core selling points in the advertisement.

[0216] Figure 5 This demonstrates the overlap of the focal areas in four advertisement images. From Figure 5 As can be seen, the overlap rates of P0, P1, P2, and P3 are 92%, 95%, 88%, and 90%, respectively, with an average of 91%. This result indicates that despite individual differences among subjects, the vast majority of people's gaze covers the core focal area, demonstrating the universal attractiveness of this region. The high overlap rate also verifies the stability and reliability of the data acquisition method—if the system has significant noise or calibration errors, the overlap rate will decrease significantly. Furthermore, this indicator excludes the interference of accidental behavior by individual subjects, proving that the analytical conclusions output by the system are representative of the population.

[0217] Figure 6 A pie chart showing the percentage of visual blind spots in four advertisement images is provided. Figure 6 (a) Figure 6 (b) Figure 6 (c) Figure 6 The middle (d) chart shows the percentage of visual blind spots for P0, P1, P2, and P3, respectively. From... Figure 6 It can be seen that the visual blind spots for P0, P1, P2, and P3 are 18%, 12%, 20%, and 15%, respectively, with an average of 16.25%. This result indicates that the overall information layout of the advertisement is relatively reasonable, and there are no large areas that are ignored. This finding provides designers with a clear direction for optimization: brand prompts or supplementary copy can be added to these areas to improve the utilization rate of advertising space.

[0218] Based on the above three indicators, the following conclusions can be drawn:

[0219] Effectiveness of the core focus area: The core area of ​​all ads accounted for more than 39% of the viewing time, indicating that the system can automatically identify and quantify the most attractive area in the ads, providing an objective basis for advertisers to verify the performance of the core selling points;

[0220] Group consistency of visual attention: The mean overlap of focal areas exceeds 90%, indicating that the eye movement data collected by the system in this paper has good inter-subject consistency and will not produce misleading conclusions due to individual random behavior. The heatmaps and indicators output by the system have high reproducibility.

[0221] The practicality of blind spot positioning: The system successfully identified 20% of the visual blind spots in P2 ads, providing designers with direct visual evidence to accurately pinpoint layout issues and optimize information distribution. By reducing the proportion of blind spots, the overall visual utilization rate can be improved without changing the core creative concept of the ad.

[0222] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for evaluating advertising effectiveness based on eye-tracking data visualization, characterized in that, Including the following steps: Configure and calibrate the eye-tracking device and display terminal to determine the pupil-screen coordinate mapping relationship; During the process of the subject naturally viewing the advertisement image, the eye-tracking device synchronously collects the gaze point coordinates, gaze duration, and pupil diameter to form an original eye-tracking dataset, wherein the gaze point coordinates are calculated in real time based on the pupil-screen coordinate mapping relationship. The raw eye-tracking data is cleaned and then normalized to obtain an effective fixation sequence; Based on the effective gaze point sequence, a gaze trajectory map is generated by linear interpolation, and a gaze heatmap is generated by Gaussian kernel density estimation. Based on the effective gaze point sequence, the gaze trajectory map, and the gaze heatmap, advertising effectiveness evaluation indicators are calculated, and a visual advertising effectiveness evaluation report is generated. The calculation of advertising effectiveness evaluation indicators and the generation of a visualized advertising effectiveness evaluation report specifically include: Aggregate the gaze heatmaps of all subjects, and automatically identify the region with the highest density as the core focus region through threshold segmentation and connected component analysis; Calculate the percentage of total fixation time in the core focus area for all subjects out of the total fixation time for the entire advertisement, as the core focus area fixation time percentage; The percentage of subjects whose gaze fell into the core focal area at least once was calculated as the focal area overlap. The area with a density value lower than the blind zone density threshold in the aggregated heat map is defined as the advertising blind zone. The proportion of pixels of this advertising blind zone to the entire advertising image is calculated as the visual blind zone proportion. A visual advertising effectiveness evaluation report is generated based on the above indicators.

2. The advertising effectiveness evaluation method based on eye-tracking data visualization according to claim 1, characterized in that, The configuration and calibration of the eye-tracking device and display terminal to determine the pupil-screen coordinate mapping relationship specifically includes: Configure an infrared pupil-corneal reflection eye-tracking device and display terminal, and set up the data acquisition environment; Multiple calibration target points are presented sequentially at preset positions on the display terminal, guiding the subject to look at each calibration target point in turn, and judging whether the calibration is passed based on the gaze deviation; Based on the relative offset vector between the pupil center and the corneal reflection point, and the pupil diameter, a multinomial regression model is used to establish a mapping function from the pupil-corneal reflection feature space to the screen pixel coordinate space, which serves as the pupil-screen coordinate mapping relationship.

3. The advertising effectiveness evaluation method based on eye-tracking data visualization according to claim 1, characterized in that, The synchronous acquisition of fixation point coordinates, fixation duration, and pupil diameter specifically includes: The advertising images to be evaluated are loaded onto the display terminal in a preset order, and then processed for size normalization, color space conversion, and brightness equalization. A central gaze point is inserted before each advertisement image is displayed to reset the initial gaze position; During the process of subjects naturally viewing the advertisement images, the pixel coordinates of the fixation point with timestamps, the duration of fixation, and the pupil diameter are collected in real time and simultaneously, and data validity labels are attached to form the original eye-tracking dataset.

4. The advertising effectiveness evaluation method based on eye-tracking data visualization according to claim 1, characterized in that, The process of cleaning and normalizing the raw eye-tracking data to obtain an effective fixation sequence specifically includes: Detect and remove blink artifacts, large head movement interference data, and abnormal data with low confidence, coordinates outside the screen range, or pupil diameter outside the physiological range; The remaining data is clustered into gaze events by using a velocity threshold method, and gaze events with a duration less than a preset threshold are filtered out to obtain a coarse gaze point sequence. The pixel coordinates in the coarse fixation point sequence are normalized to a unit interval according to the screen width and height to obtain the effective fixation point sequence.

5. The advertising effectiveness evaluation method based on eye-tracking data visualization according to claim 4, characterized in that, The detection and removal of blink artifact data specifically includes: jointly determining blink events based on the rate of change of pupil area between adjacent frames and the abrupt change in the position of corneal reflective points, and removing all eye movement data within a preset time window before and after the blink event.

6. The advertising effectiveness evaluation method based on eye-tracking data visualization according to claim 4, characterized in that, The detection and removal of large head movement interference data specifically includes: acquiring head displacement velocity and rotation angular velocity; when the displacement velocity or rotation angular velocity exceeds the corresponding threshold, it is determined as a head interference event, and all eye movement data during the duration of the interference event and within a preset time before and after it are removed.

7. The advertising effectiveness evaluation method based on eye-tracking data visualization according to claim 1, characterized in that, The generation of the gaze trajectory map through linear interpolation specifically includes: The gaze events are extracted from the effective gaze sequence and sorted in ascending order by timestamp to obtain an ordered gaze sequence. The spatial location of each gaze event is used as a trajectory node; Multiple intermediate points are generated by uniformly interpolating over time between adjacent trajectory nodes. The points are connected to form an interpolation path. The trajectory nodes and the interpolation path are then overlaid on the original advertisement image to form a gaze trajectory map.

8. The advertising effectiveness evaluation method based on eye-tracking data visualization according to claim 1, characterized in that, The generation of the gaze heatmap through Gaussian kernel density estimation specifically includes: Using the normalized coordinates of each fixation event in the effective fixation point sequence as the center and the fixation duration as the weight, the attention density value of each point on the normalized plane is calculated by a two-dimensional Gaussian kernel function. The normalized plane is discretized into a grid, and the density value of each grid cell is calculated to obtain a two-dimensional density matrix; After linearly normalizing the original density values ​​in the two-dimensional density matrix, they are converted into color values ​​according to a preset color mapping table to generate a heat map layer, which is then superimposed on the original advertising image in a semi-transparent manner to form a gaze heat map.

9. An advertising effectiveness evaluation system based on eye-tracking data visualization, characterized in that, include: The experimental configuration and calibration module is used to configure the eye-tracking device and display terminal and perform calibration to establish the pupil-screen coordinate mapping relationship. The data acquisition module is used to simultaneously collect fixation point coordinates, fixation duration, and pupil diameter while subjects view advertising images, forming a raw eye-tracking dataset; The data cleaning and gaze detection module is used to clean the raw eye movement data, extract gaze events and normalize coordinates, and output a valid gaze point sequence. The heatmap and trajectory map generation module is used to generate gaze trajectory maps based on effective gaze point sequences through linear interpolation and to generate gaze heatmaps through Gaussian kernel density estimation. The metrics calculation and reporting module is used to calculate advertising effectiveness evaluation metrics based on effective fixation point sequences, fixation trajectory maps, and fixation heatmaps, and generate visual evaluation reports. The calculation of advertising effectiveness evaluation indicators and the generation of a visualized advertising effectiveness evaluation report specifically include: Aggregate the gaze heatmaps of all subjects, and automatically identify the region with the highest density as the core focus region through threshold segmentation and connected component analysis; Calculate the percentage of total fixation time in the core focus area for all subjects out of the total fixation time for the entire advertisement, as the core focus area fixation time percentage; The percentage of subjects whose gaze fell into the core focal area at least once was calculated as the focal area overlap. The area with a density value lower than the blind zone density threshold in the aggregated heat map is defined as the advertising blind zone. The proportion of pixels of this advertising blind zone to the entire advertising image is calculated as the visual blind zone proportion. A visual advertising effectiveness evaluation report is generated based on the above indicators.

Images