Indoor illumination simulation and optimization system and method based on AI

Abstract

The invention discloses an AI-based indoor illumination simulation and optimization system and method, and relates to the technical field of illumination processing.The method comprises the steps that firstly, user skeleton key point coordinates, screen light reflection area coordinates and sight line focus coordinates are obtained in real time, and the displacement of a light reflection area is calculated based on the head pitch angle and the shoulder inclination angle in combination with a preset linear displacement coefficient; and generating dynamic prediction coordinates. By calculating the overlapping ratio of the sight focus and the predicted reflective area, whether an optimization instruction is triggered or not is intelligently judged, and when the threshold value is exceeded, the brightness and the inclination angle of the associated lamp are automatically adjusted to ensure that the brightness of the actual reflective area is reduced below the safety threshold value. And finally, calculating a drop amplitude by re-collecting a brightness value, and feeding back and adjusting a linear displacement coefficient to realize closed-loop optimization. The problem of dynamic glare caused by posture change of a user in an indoor illumination environment is effectively solved, and the visual comfort and the working efficiency are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lighting processing, in particular to an AI-based indoor lighting simulation and optimization system and method thereof. BACKGROUND

[0002] In modern architectural design and indoor environment creation, the quality and effect of indoor lighting have a crucial influence on people's visual experience, work efficiency, and physical and mental health. Reasonable indoor lighting not only can provide sufficient illumination and reduce visual fatigue, but also can create a comfortable and pleasant space atmosphere. In office, education, medical and other scenarios, when users face electronic screens for a long time, screen reflection (glare) is an important factor leading to visual fatigue and distraction.

[0003] The prior art mainly adopts the following ways to alleviate the glare problem, but still has many deficiencies: static shading scheme: such as anti-glare screen stickers, fixed shading boards, etc., which cannot adapt to the dynamic changes of user posture and lighting conditions, and have poor flexibility. Ambient light sensor adjustment: only relying on ambient light intensity to adjust the brightness of the lamp, which cannot accurately identify the screen reflection area, and does not consider the user's visual focus, which may lead to excessive dimming or ineffective adjustment. Glare detection based on camera: some schemes detect screen reflection through cameras, but do not combine user head posture and visual direction, resulting in large prediction error and limited optimization effect. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides an AI-based indoor lighting simulation and optimization system and method thereof.

[0005] In order to achieve the above purpose, the technical scheme of the present application is as follows:

[0006] In a first aspect, the present application discloses an AI-based indoor lighting simulation and optimization method, comprising the following steps:

[0007] Obtain the user's skeletal key point coordinates, screen reflection area coordinates and visual focus coordinates;

[0008] Based on the head pitch angle and shoulder tilt angle of the skeletal key point coordinates, calculate the displacement amount of the reflection area through the preset linear displacement coefficient, and generate dynamic reflection area prediction coordinates;

[0009] Calculate the overlap rate of the visual focus coordinates and the dynamic reflection area prediction coordinates, and generate an optimization instruction when the overlap rate exceeds a preset overlap rate threshold;

[0010] Adjust the brightness and tilt angle of the preset associated lamps according to the optimization instruction, so that the actual reflection area brightness after adjustment falls below the preset safe brightness threshold;

[0011] The actual reflection area brightness value is re-acquired, the brightness drop amplitude is calculated, and the linear displacement coefficient is adjusted in proportion according to the difference between the brightness drop amplitude and a preset target value.

[0012] In a second aspect, the present application discloses an AI-based indoor lighting simulation and optimization system, which uses the AI-based indoor lighting simulation and optimization method described above, and comprises:

[0013] A data acquisition module is configured to acquire the user's skeletal key point coordinates, screen reflection area coordinates, and line-of-sight focus point coordinates.

[0014] A dynamic prediction module is configured to calculate the displacement amount of the reflection area based on the offset angle of the skeletal key point coordinates through a preset linear coefficient to generate dynamic reflection area prediction coordinates.

[0015] An overlap determination module is configured to calculate the overlap rate of the line-of-sight focus point coordinates and the dynamic reflection area prediction coordinates, and generate an optimization instruction when the overlap rate exceeds a preset overlap rate threshold.

[0016] A light source control module is configured to adjust the brightness and tilt angle of the preset associated lamps according to the optimization instruction, so that the adjusted actual reflection area brightness drops below a preset safe brightness threshold.

[0017] A verification feedback module is configured to re-acquire the actual reflection area brightness value after adjusting the lamp parameters, calculate the brightness drop amplitude, and adjust the linear displacement coefficient in proportion according to the difference between the brightness drop amplitude and a preset target value.

[0018] A historical data management module is configured to statistically analyze the historical success rate of each parameter combination in the historical optimization database, and preferentially select the parameter combination that has been verified to be effective in the historical optimization database based on the lamp parameter adjustment records of the same skeletal key point coordinates to execute adjustment.

[0019] Compared with the prior art, the present application has the following advantages:

[0020] 1. The displacement amount of the reflection area is calculated based on the skeletal key points to achieve dynamic prediction, and the prediction accuracy is improved by combining the calibration and feedback adjustment of the linear displacement coefficient, thereby reducing the prediction error caused by the change of the user's posture and making the reflection area tracking more accurate.

[0021] 2. The overlap rate of the line-of-sight focus point coordinates and the dynamic reflection area is calculated, and the optimization instruction is triggered when the overlap rate exceeds the threshold, and the optimization threshold is dynamically adjusted in combination with the user fatigue data to achieve personalized adjustment, avoid invalid dimming of non-attention areas, optimize resource allocation, and improve energy efficiency.

[0022] 3. Calculate the brightness contribution of each luminaire to the light reflection area, normalize it as weight, and through multi-luminaire cooperative control, accurately reduce the brightness of the target area while maintaining overall illumination balance. BRIEF DESCRIPTION OF DRAWINGS

[0023] The disclosure of the present application will be described with reference to the accompanying drawings. It should be understood that the drawings are only for illustrative purposes, and are not intended to limit the scope of protection of the present application. In the drawings, the same reference numerals are used to refer to the same components. Among them:

[0024] Figure 1 is a specific flowchart of step one of the present application;

[0025] Figure 2 is a specific flowchart of step one of the present application;

[0026] Figure 3 is a process chart of closed-loop calibration of the present application;

[0027] Figure 4 is a real-time analysis flowchart of the brightness reduction amplitude of the present application;

[0028] Figure 5 is a parameter combination intelligent optimization flowchart based on the history optimization database of the present application. DETAILED DESCRIPTION

[0029] It is easy to understand that according to the technical scheme of the present application, those skilled in the art can propose a plurality of structure modes and implementation modes which can be replaced with each other without changing the essential spirit of the present application. Therefore, the following specific embodiments and drawings are only exemplary description of the technical scheme of the present application, and should not be regarded as the whole or as the limitation or restriction of the technical scheme of the present application.

[0030] SUMMARY

[0031] As mentioned above, with the rapid development of artificial intelligence (AI) technology, its application in various fields is becoming more and more widespread. In the field of indoor lighting, although there have been some attempts to apply AI technology to lighting simulation and optimization, it is still in its infancy. The existing AI-based indoor lighting scheme mainly focuses on the macro analysis and prediction of the overall lighting environment, and lacks in-depth research on individual user behavior and visual experience. For example, the interaction between user's skeletal key points, visual focus and indoor lighting is not fully considered, and fine lighting simulation and optimization cannot be achieved. In addition, the adaptability and real-time performance of existing technologies in handling dynamic changes in indoor lighting scenarios need to be improved, and it is difficult to meet the complex demands in actual applications.

[0032] In view of the above defects in the prior art, the basic idea of the present application is to first obtain user bone key point coordinates, screen reflection area coordinates and line of sight focus point coordinates in real time, calculate the reflection area displacement amount based on the head pitch angle and the shoulder tilt angle in combination with a preset linear displacement coefficient, and generate dynamic prediction coordinates. By calculating the overlap rate of the line of sight focus point and the predicted reflection area, it is intelligently judged whether to trigger the optimization instruction, and when the threshold is exceeded, the brightness and tilt angle of the associated lamps are automatically adjusted to ensure that the actual reflection area brightness is reduced below the safety threshold. Finally, the linear displacement coefficient is adjusted by re-acquiring the brightness value to calculate the drop amplitude to realize closed-loop optimization.

[0033] After introducing the basic idea of the present application, the embodiments of the present application will be specifically introduced below with reference to the drawings.

[0034] Embodiments

[0035] As shown in Figure 1 , the AI-based indoor lighting simulation and optimization method comprises the following steps:

[0036] Step one, obtain the user's bone key point coordinates, screen reflection area coordinates and line of sight focus point coordinates.

[0037] The bone key point coordinates represent the key coordinates of the user's torso, which can reflect the user's sitting or standing posture, such as whether to lean forward or lean to one side; for example, if the head center point is detected to move forward + the spine midpoint is convex upward, it indicates that the user is leaning forward to look at the screen, and at this time the reflection area may move upward.

[0038] The screen reflection area coordinates are represented as the position information of the high-brightness reflection area generated on the surface of the electronic screen due to the environmental light irradiation, which is usually represented in pixel coordinates or physical coordinates; it can reflect the specular reflection distribution of the environmental light source on the screen surface, and identify the strong light area that may cause visual discomfort.

[0039] The line of sight focus point coordinates are the screen position coordinates actually gazed by the user's eyeballs, which are the visual attention center points obtained through line of sight tracking technology; it can reflect the priority area of brain visual information processing, and the core area of current information acquisition of the user.

[0040] As shown in Figure 2 , the specific process of step one is as follows:

[0041] The binocular camera captures the bone key point coordinates of the user's upper body at a preset sampling frequency, wherein the bone key point coordinates of the user's upper body include: head center point, left and right shoulder points, spine upper, middle and lower points; the head pitch angle a is calculated by the included angle of the line connecting the spine midpoint and the head center point, and the shoulder tilt angle β is calculated by the included angle of the line connecting the left and right shoulder points and the horizontal plane;

[0042] Head center: the center of the user's head or face, used to calculate the head pose.

[0043] Left and right shoulder points: the positions of the two shoulder joints, used to determine the shoulder tilt.

[0044] Upper, middle, and lower spine points: key positions on the spine, used to assist in calculating the body pose.

[0045] Head tilt angle α is calculated by the angle between the line connecting the middle spine point and the head center, reflecting the up-and-down tilt of the head (looking up / looking down). When the angle α > 0°, the user is looking up, and the reflective area moves down. When α < 0°, the user is looking down, and the reflective area moves up.

[0046] Shoulder tilt angle β is calculated by the angle between the line connecting the left and right shoulder points and the horizontal plane, reflecting the left-and-right tilt of the user's shoulders. When the angle β ≠ 0°, it indicates that the user's body is leaning to one side (e.g., leaning against the armrest of a chair), which will cause the screen's reflective area to shift horizontally.

[0047] Based on the RGB pixel values of the screen surface, the coordinates of the continuous area with a brightness value exceeding a preset reflective light threshold are extracted.

[0048] Using a high dynamic range camera to capture the screen surface image, the image data is converted from RGB to HSV color space, and then the brightness channel is extracted. An adaptive threshold algorithm (Otsu method) is used to determine the reflective light threshold, and the threshold is dynamically adjusted in combination with the ambient light sensor data. After processing, the continuous reflective area is marked. This method can identify the actual reflective area on the screen surface, avoid energy waste caused by global dimming, adapt to changes in ambient light through real-time image processing, and exclude false judgments caused by high-brightness elements in the screen content.

[0049] The gaze focus coordinates in the polar coordinate system with the screen center as the origin are obtained through an infrared eye tracking device.

[0050] The pupil center and corneal reflection point are captured by an infrared camera, the gaze vector is calculated, and the gaze focus coordinates are obtained after coordinate system conversion.

[0051] Step two: Based on the head tilt angle and shoulder tilt angle calculated from the skeletal key point coordinates, the displacement of the reflective area is calculated through a pre-set linear displacement coefficient, and the dynamic reflective area prediction coordinates are generated.

[0052] This process is the core calculation link of the light simulation adjustment. By converting the user's real-time pose data (head / shoulder angles) into displacement prediction of the screen's reflective area, the following effects are achieved: solving the hysteresis problem of traditional static detection (traditional methods cannot adapt to the changes in reflection caused by human movement by fixing the detection area), achieving predictive adjustment (calculating the possible reflective area in advance and adjusting the light in advance before the user actually feels glare).

[0053] Geometric features are extracted from the skeletal key points (such as the head center, shoulder points, and spine points) obtained by the camera, including the head pitch angle and the shoulder tilt angle, and a linear mapping relationship between the angle change and the pixel displacement is established. According to the mapping relationship, the displacement amount is added to the screen reflection area coordinates obtained in step one, and the predicted coordinates are output. Only simple trigonometric functions and linear transformations need to be calculated, which can be completed within 10 ms. By using skeletal key points instead of direct image recognition, the interference of environmental light changes is avoided.

[0054] The displacement amount of the reflection area includes:

[0055] According to the head pitch angle α and the shoulder tilt angle β, the displacement amount is calculated as horizontal displacement Δx = k1 × β and vertical displacement Δy = k2 × α, where k1 and k2 are pre-set linear displacement coefficients.

[0056] The meanings of k1 and k2 are respectively the pixel displacement amount corresponding to each degree of angle change, and the head pitch angle α and the shoulder tilt angle β represent the user's body state, respectively. The angles have a linear-like relationship with the vertical displacement Δy and the horizontal displacement Δx within a certain range. In this way, the head pitch angle α and the shoulder tilt angle β are calculated to calculate the horizontal displacement and the vertical displacement.

[0057] The reflection area coordinates are translated along the X-axis by Δx and along the Y-axis by Δy to generate dynamic reflection area predicted coordinates.

[0058] Example:

[0059] Input data acquisition:

[0060] The user's skeletal key points are captured by the camera, and the real-time angles are calculated: head pitch angle α = -12° (user bends down) and shoulder tilt angle β = 8° (right shoulder sinks).

[0061] Displacement amount calculation:

[0062] If k1 = 1.2 and k2 = 0.8, then:

[0063] Horizontal displacement: Δx = k1 × β = 1.2 × 8 = 9.6 pixels;

[0064] Vertical displacement: Δy = k2 × α = 0.8 × (-12) = -9.6 pixels.

[0065] Coordinate dynamic prediction:

[0066] Original reflection area coordinates (x0, y0) → predicted coordinates (x0 + 9.6, y0 - 9.6).

[0067] The linear displacement coefficients k1 and k2 need to be calibrated to ensure that their values are reasonable, and thus ensure that the displacement of the reflective area calculated based on the linear displacement coefficients can reflect the changes caused by the change in the human body posture, thereby ensuring the accuracy of the results.

[0068] The calibration process is as follows:

[0069] In the initialization phase, guide the user to perform a standard posture action;

[0070] The specific process of performing a standard posture action is as follows:

[0071] First, guide the user to maintain a standard sitting posture (head straight up, α = 0°, shoulders horizontal, β = 0°), and record the reference coordinates of the screen reflective area (x ref ,y ref ) at this time;

[0072] Then guide the user to perform different actions;

[0073] For example: pure head-down action, at this time the shoulder does not move β' = 0°, the head-down angle α' = -15°, and the actual reflective area vertical displacement Δy' is measured;

[0074] Pure shoulder tilt action, at this time the head does not move α' = 0°, the shoulder tilt angle β' = 10°, and the actual reflective area horizontal displacement Δx' is measured;

[0075] Based on the ratio of the actual reflective area displacement Δx', Δy' to the measured angle β', α', update k1 = Δx' / β', k2 = Δy' / α'.

[0076] For example: If head-down α' = -15° causes the reflective area to move up Δy' = 12 pixels → k2 = 12 / 15 = 0.8 pixels / degree.

[0077] If the shoulder tilt β' = 10° causes the reflective area to move Δx' = 12 pixels → k1 = 12 / 10 = 1.2 pixels / degree.

[0078] Different users have different reflective displacement rules due to differences in height and sitting habits, and need to be adjusted adaptively according to use.

[0079] Step three, calculate the overlap rate of the line of sight focus coordinates and the dynamic reflective area prediction coordinates, and generate an optimization instruction when the overlap rate exceeds a preset overlap rate threshold.

[0080] The overlap rate of the line of sight focus coordinates and the dynamic reflective area prediction coordinates is used to quantify the spatial overlap degree of the user's line of sight focus and the predicted reflective area, and thus the following goals are achieved:

[0081] Determine the glare risk: avoid false positives due to proximity only, introduce physical plausibility of area overlap.

[0082] Dynamic threshold response: balance sensitivity (false negatives) and stability (false positives) by presetting an overlap rate threshold (e.g. 70%).

[0083] Trigger closed-loop control: provide clear execution basis for subsequent light adjustment (step four) and effect feedback (step five).

[0084] Obtain the user's current gaze point and its impact range (e.g. a circular area with a radius r = 30 pixels) through eye tracking or head pose estimation in step one;

[0085] Predict the coordinates of the dynamic reflective area through prediction in step two.

[0086] Use the geometric area intersection ratio method, the formula is:

[0087]

[0088] The preset overlap rate threshold needs to be calibrated through user experience to balance the need for glare suppression and operational fluency; by setting multiple threshold values, the overlap of the overlap rate is classified, and different types of optimization measures are taken to achieve the effect of hierarchical adjustment, thereby improving the optimization effect.

[0089] Calculating the overlap rate of the line of sight focus point coordinates and the dynamic reflective area prediction coordinates includes:

[0090] Expand the line of sight focus point coordinates into a circular area with a preset radius centered on the focus point;

[0091] The preset radius circular area can meet the physiological characteristics of the human eye, and the circular area covers the main clear vision range.

[0092] For example: if the line of sight focus point is (300, 200) and the radius r = 20;

[0093] Then the covered pixel coordinates satisfy (x-300) 2 +(y-200) 2 ≤400.

[0094] Calculate the geometric overlap area ratio of the preset radius circular area and the dynamic reflective area prediction coordinates as the overlap rate;

[0095] The overlap rate can reflect the interference of the reflective area on the user's viewing of the screen, and by quantifying the geometric probability of visual interference, it provides a quantifiable decision basis for the intelligent lighting system, and is a key bridge connecting the user's physiological state and hardware control.

[0096] As shown in the following table, some overlap rate ranges correspond to physiological explanations:

[0097] Overlap rate range Physiological explanation 0～30％ The reflective area is located at the edge of the visual field, where the sensitivity of the human eye's peripheral retina is low 30％～70％ The reflection enters the area with high density of cone cells, which may cause slight visual fatigue >70％ The reflection covers the fovea, directly interfering with the acquisition of visual information

[0098] When the overlap rate exceeds the first-level optimization threshold and the average brightness of the reflective area is >150, generate a first-level optimization instruction;

[0099] When the overlap rate is between the second-level optimization threshold and the first-level optimization threshold, generate a second-level optimization instruction; the first / second-level instructions correspond to different optimization effects.

[0100] The first-level optimization threshold and the second-level optimization threshold need to be calibrated through experiments in actual applications. The brightness threshold of 150 corresponds to an approximate 60 nit brightness, which is a critical value for medium and high glare risk.

[0101] The first-level optimization threshold is greater than the second-level optimization threshold. The first-level optimization instruction corresponding to the first-level optimization threshold is used to process direct glare risk and corresponds to rigid intervention. The second-level optimization instruction corresponding to the second-level optimization threshold is used to process potential glare risk and corresponds to soft intervention.

[0102] The first-level optimization threshold and the second-level optimization threshold are linearly reduced according to user fatigue data, which is the length of time that the user continuously gazes at the screen.

[0103] Long gazing time can cause unstable tear film and increase the glare sensitivity by 30% to 50%. Therefore, the first-level optimization threshold and the second-level optimization threshold need to be reduced to lower the triggering standard of the optimization instruction and make it easier to be triggered for optimization.

[0104] The calculation model is:

[0105] The current threshold = initial threshold × max[0.5, 1-0.01 × gazing time (minutes)]. The limit value of 0.5 is set to avoid unlimited reduction of the threshold.

[0106] Compared with the traditional single threshold, the combination of the geometric overlap rate, the area brightness, and the user fatigue data is more scientific. The threshold is reduced in real time according to the gazing time, which meets the glare sensitivity characteristics of the human eye after fatigue.

[0107] Example:

[0108] Scene setting:

[0109] Screen resolution: 1920×1080;

[0110] Line of sight focal point coordinates: (500, 300);

[0111] Dynamic reflective area: center coordinates (520, 310), radius R2 = 40 pixels;

[0112] Preset parameters:

[0113] Line of sight focus area radius R1 = 30 pixels;

[0114] Initial primary threshold is 70%, secondary threshold is 40%;

[0115] The user has been continuously gazing at the screen for 25 minutes.

[0116] Calculate the center distance:

[0117]

[0118] Overlap area calculation:

[0119] Use the fast approximation formula:

[0120]

[0121] Overlap rate result:

[0122]

[0123] After considering the user fatigue data, the threshold adjustment result is:

[0124] Primary threshold: 70 × max[0.5, 1-0.01 × 25] = 70 × 0.75 = 52.5;

[0125] Secondary threshold: 40 × max[0.5, 1-0.01 × 25] = 40 × 0.75 = 30.

[0126] The overlap rate does not reach the threshold, and the primary and secondary instructions are not triggered, and the current light state is maintained.

[0127] Change scenario simulation:

[0128] Assume that the reflection area moves to (490, 305),

[0129] New center distance:

[0130] New overlap rate ≈ 58.7%;

[0131] The new overlap rate is greater than the primary threshold, and the condition that the average brightness of the reflection area is greater than 150 is met, triggering the primary instruction to optimize.

[0132] Step four, adjust the brightness and inclination angle of the preset associated lamps according to the optimization instruction, so that the adjusted actual reflection area brightness drops below the preset safety brightness threshold.

[0133] This step is the implementation link, according to the optimization instruction of step three, take corresponding adjustment action, used to convert the calculation result of the previous step into physical device action, realize:

[0134] Light intensity control: Adjust the brightness / angle of the lamp to reduce the brightness of the screen reflection area to the comfortable range of the human eye.

[0135] Spatial position compensation: Adjust the tilt angle to change the reflection path and avoid the user's visual focus area.

[0136] Closed-loop verification: Ensure that the actual brightness after adjustment meets the preset safety threshold, otherwise trigger the feedback calibration process in step five.

[0137] Adjusting the brightness and tilt angle of the preset associated lamps according to the optimization instruction includes:

[0138] Calculate the brightness contribution of each lamp to the reflection area: Where C i represents the brightness contribution of the i-th lamp to the reflection area, L i is the lamp brightness of the i-th lamp, θ i is the angle between the lamp light and the normal of the reflection area, and d i is the straight-line distance from the lamp to the reflection area.

[0139] The brightness contribution is calculated based on the law of optical propagation (inverse square + cosine attenuation), which is more scientific than empirical adjustment.

[0140] Example:

[0141] If the lamp brightness is 300 nit, the angle is 30°, and the distance is 1.5m:

[0142]

[0143] Normalize the brightness contribution of each lamp to the reflection area to the weight of each lamp

[0144] The purpose of normalization is to convert the contribution of each lamp to the weight in the interval [0,1], reflecting its relative influence. The denominator is the sum of the contributions of all associated lamps, ensuring that ∑W i =1.

[0145] Example: If the contribution of three groups of lamps is [115.5, 80.2, 45.3], the weight of the first group of lamps is:

[0146]

[0147] In response to the first-level optimization instruction, adjust the lamp with the highest weight, and its brightness is reduced to:

[0148] L adjusted = L i ·(1-W i ·0.5).

[0149] The greater the contribution and the higher the weight, the greater the brightness attenuation ratio, up to 50%.

[0150] Example: if W i = 0.48, the brightness is reduced to L i × (1-0.48×0.5) = 0.76L i .

[0151] The tilt angle adjustment amount is:

[0152]

[0153] The light with the highest weight gets the maximum angle adjustment, i.e., the adjustment angle is 15°, and the rest are scaled proportionally;

[0154] Example: if W i = 0.48 is the maximum value, its adjustment angle is 15°; if another light W j = 0.32, then the adjustment is 15°×0.32 / 0.48 = 10°.

[0155] In response to the secondary optimization instruction, the color temperature of the light with the top two weights is adjusted synchronously to:

[0156] CCT adjusted = CCT i -(W i ·500), and no less than 2500K.

[0157] CCT i is the initial color temperature of the light, W i ·500 is the color temperature attenuation amount, with a maximum of 500K per time;

[0158] Example:

[0159] If the initial color temperature of the light is 5500K and the weight is 0.3:

[0160] CCT adjusted = 5500-(0.3×500) = 5350K;

[0161] If the weight is 0.8:

[0162] CCT adjusted = 5500-(0.8×500) = 5100K.

[0163] The primary instruction prioritizes the main interference source (the light with the highest weight) to reduce unnecessary energy consumption; the secondary instruction compensates for the color temperature of multiple lights, with a lower limit (≥2500K) for color temperature adjustment to avoid excessive coldness leading to discomfort.

[0164] Example:

[0165] Scenario: Three-lamp system, level-1 instruction trigger

[0166] Lamp parameters are shown in the following table:

[0167] Lamp ID Brightness (nit) Color temperature (K) Distance (m) Angle (°) 1 300 5500 1.5 30 2 250 5500 1.2 45 3 200 5500 2.0 15

[0168] Calculation process:

[0169] Contribution calculation:

[0170]

[0171] Weight normalization:

[0172] (Highest weight).

[0173] Adjustment execution:

[0174] Brightness: L2 = 250 x (1 - 0.43 x 0.5) ≈ 196 nit

[0175] Tilt angle:

[0176] When it is detected that the user continuously gazes at the same area for more than a preset time threshold, the weight of the associated lamp is increased by 50%.

[0177] The preset time threshold is usually set to 30 seconds, which is determined based on the stability of the human eye tear film. Gazing for more than 30 seconds can easily lead to dry eye syndrome; the logic for weight increase is:

[0178] W new = min(1.0, W current x 1.5), setting the upper limit value of the increased weight to 1.0 to avoid excessive illumination; the associated lamps represent the three nearest lamps mapped through the spatial topology table.

[0179] When it is detected that the user's head rotation angle exceeds a preset angle threshold, the weight of the associated lamp in the new gazing area is reset to the initial value.

[0180] The preset angle threshold has two sets, corresponding to horizontal rotation and vertical rotation, respectively. The threshold values for horizontal rotation and vertical rotation correspond to screen edge jumping, such as the horizontal rotation from the center to the side menu bar.

[0181] By responding to the user's gazing behavior in real time, the dynamic optimization of the lamp weight (brightness / importance) is achieved.

[0182] Step five, reacquire the actual reflected area brightness value, calculate the brightness drop amplitude, and adjust the linear displacement coefficient in proportion according to the difference between the brightness drop amplitude and the preset target value.

[0183] For example,Figure 3 As shown, this step is a closed-loop calibration process that reverses the optimization of the prediction model by measuring the brightness changes in real time to achieve:

[0184] Error compensation: eliminate the prediction deviation caused by factors such as sudden changes in ambient light and screen aging;

[0185] Adaptive learning: make the linear displacement coefficients (k1, k2) evolve dynamically with the use of the scene;

[0186] Stability guarantee: prevent the system from entering an uncontrollable state due to a single misjudgment.

[0187] The adjusted actual light reflection area brightness value is collected by the sensor, and the brightness drop amplitude is obtained by combining it with the light reflection area brightness value before adjustment:

[0188] Example: if the initial brightness is 120 nit and the current brightness is 83.6 nit:

[0189]

[0190] The process of adjusting the linear displacement coefficient in proportion is:

[0191] Get the brightness drop amplitude ΔL;

[0192] If ΔL ≥ 15%, then:

[0193]

[0194] Take 20% as the critical point, and increase the coefficient by (ΔL-20%) for every 1% of the brightness drop amplitude ΔL;

[0195] For example: ΔL = 25% → coefficient amplification 5% (1.05 times);

[0196] When ΔL > 20%, it means that the current coefficient has underestimated the actual demand and needs to increase the sensitivity;

[0197] When 15% ≤ ΔL < 20%, slightly reduce the coefficient (e.g. ΔL = 18% → coefficient × 0.98).

[0198] And when ΔL < 15%, mark the current lamp parameter combination as an inefficient parameter and reduce its historical weight to 0.2, skip such optimization in the future, and reduce the priority of this combination in AI recommendation.

[0199] Through multi-level processing of brightness change data, dynamic calibration of parameters is realized, linear displacement coefficients (k1, k2) are corrected in reverse according to the actual brightness drop amplitude ΔL, and the prediction accuracy is improved; when ΔL≥15%, it is determined that the optimization combination is effective, and is stored in the database for future learning, and when ΔL<15%, the parameter is marked as low efficiency; the historical optimization database provides real scene data for subsequent AI training.

[0200] As shown in Figure 4 , it is a real-time analysis flowchart of brightness drop amplitude;

[0201] When the brightness drop amplitude is less than 20%, the physical light shield control instruction is triggered;

[0202] 20% corresponds to the critical value of glare perception, and when the electronic dimming reaches the limit (such as the lowest brightness of the lamp), the physical light shield is the final guarantee.

[0203] When the brightness drop amplitude is greater than or equal to 20%, the current skeleton key point coordinates, displacement amount and lamp parameter combination are stored in the historical optimization database.

[0204] Only the effective data of ΔL≥20% is stored in the database to avoid noise pollution, which is used to compare the optimal parameter combination under similar postures in real time, train the AI model to predict more accurate linear displacement coefficients (k1, k2), and improve the prediction accuracy.

[0205] Example:

[0206] Scenario 1: effective optimization (ΔL=25%);

[0207] Original coefficients: k1=1.2, k2=0.8;

[0208] New coefficient calculation:

[0209]

[0210] Scenario 2: low efficiency response (ΔL=10%);

[0211] Action: mark the current lamp parameter combination as low efficiency.

[0212] After generating the optimization instruction, it also includes:

[0213] Based on the lamp parameter adjustment records of the same skeleton key point coordinates in the historical optimization database, the parameter combination verified effectively in history is preferentially selected to perform adjustment;

[0214] By calling the parameter combination verified effectively in history, the following is achieved:

[0215] Accurate decision: avoid repeated calculation and directly apply the verified efficient scheme;

[0216] Efficiency improvement: reduce real-time operation load, response speed improved by more than 40%.

[0217] As Figure 5 shown, the parameter combination intelligent optimization flowchart based on historical optimization database;

[0218] The historical optimization database stores the combination of bone key point pixel coordinates, displacement (Δx, Δy), brightness adjustment percentage, lamp tilt angle, and historical success rate. Index mapping is established for bone key point pixel coordinates to accelerate similar posture query. When there is no matching record, fall back to the default parameters and mark it as a new case.

[0219] The process of preferentially selecting the parameter combination verified effective by history to execute adjustment includes:

[0220] Statistical history success rate of each parameter combination Generate priority weight W priority ;

[0221] N success represents the number of times the combination successfully suppresses glare, i.e. ΔL≥20% after the combination is executed, N total is the total number of times the combination is called; after each execution, update N success and N total according to whether ΔL≥20%.

[0222] Example:

[0223] If P001 combination is successful 28 times / total 30 times→W priority =28 / 30≈0.933.

[0224] When the same bone key point coordinates are newly detected, preferentially call the parameter combination with W priority >0.8.

[0225] W priority >0.8 means that the success rate is more than 80%, which is the reliability threshold verified by experiments.

[0226] Example:

[0227] Scenario: user reading (bone key point matching P001 combination);

[0228] Data matching:

[0229] The current posture matches P001, and the weight of P001 is 0.933 (>0.8).

[0230] Parameter call:

[0231] Directly apply the lamp settings of P001: brightness reduction by 40%, tilt by -10°.

[0232] Effect verification:

[0233] Measured Delta L = 22% -> update N of P001 success = 29, N total = 31.

[0234] In the second embodiment, the AI-based indoor lighting simulation and optimization system is disclosed, which uses the AI-based indoor lighting simulation and optimization method described above, comprising:

[0235] A data acquisition module is configured to acquire the user's skeletal key point coordinates, screen reflection area coordinates and line of sight focus point coordinates.

[0236] A dynamic prediction module is configured to calculate the displacement of the reflection area based on the offset angle of the skeletal key point coordinates through a preset linear coefficient to generate dynamic reflection area prediction coordinates.

[0237] An overlap determination module is configured to calculate the overlap rate of the line of sight focus point coordinates and the dynamic reflection area prediction coordinates, and generate an optimization instruction when the overlap rate exceeds a preset overlap rate threshold.

[0238] A light source control module is configured to adjust the brightness and tilt angle of the preset associated lamps according to the optimization instruction, so that the adjusted actual reflection area brightness decreases below the preset safety brightness threshold.

[0239] A verification feedback module is configured to reacquire the actual reflection area brightness value after adjusting the lamp parameters, calculate the brightness drop amplitude and adjust the linear displacement coefficient in proportion according to the difference between the brightness drop amplitude and the preset target value.

[0240] A historical data management module is configured to statistically analyze the historical success rate of each parameter combination in the historical optimization database, and based on the lamp parameter adjustment records of the same skeletal key point coordinates in the historical optimization database, preferentially select the historical verification effective parameter combination to execute adjustment.

[0241] The technical scope of the present application is not limited to the content in the above description, and those skilled in the art can make various modifications and changes to the above embodiments without departing from the technical idea of the present application, and these modifications and changes should all belong to the protection scope of the present application.

Claims

1. AI-based indoor lighting simulation and optimization method, characterized in that: The method comprises the following steps: Obtaining the skeletal key point coordinates of the user, the screen reflection area coordinates and the line of sight focus point coordinates; Based on the head pitch angle and the shoulder tilt angle of the skeletal key point coordinates, the displacement amount of the reflection area is calculated through the preset linear displacement coefficient to generate the dynamic reflection area prediction coordinates; The overlap rate of the line of sight focus point coordinates and the dynamic reflection area prediction coordinates is calculated, and an optimization instruction is generated when the overlap rate exceeds the preset overlap rate threshold; According to the optimization instruction, the brightness and tilt angle of the preset associated lamps are adjusted, so that the adjusted actual reflection area brightness decreases to below the preset safety brightness threshold; The actual reflection area brightness value is re-acquired, the brightness decrease amplitude is calculated, and the linear displacement coefficient is adjusted in proportion according to the difference between the brightness decrease amplitude and the preset target value. 2.The AI-based indoor lighting simulation and optimization method of claim 1, wherein: The method comprises the following steps: The skeletal key point coordinates of the user's upper body are captured by the binocular camera at a preset sampling frequency, and the skeletal key point coordinates of the user's upper body include the head center point, the left and right shoulder points, and the upper, middle and lower points of the spine; the head pitch angle α is calculated by the included angle of the line connecting the middle point of the spine and the head center point, and the shoulder tilt angle β is calculated by the included angle of the line connecting the left and right shoulder points and the horizontal plane; Based on the screen surface RGB pixel value, the continuous area coordinates with the brightness value exceeding the preset reflection brightness threshold are extracted; The line of sight focus point coordinates in the polar coordinate system with the screen center as the origin are obtained by the infrared eyeball tracking device. 3.The AI-based indoor lighting simulation and optimization method of claim 1, wherein: The method comprises the following steps: According to the head pitch angle α and the shoulder tilt angle β, the displacement amount is calculated according to Δx=k1×β and Δy=k2×α, wherein k1 and k2 are preset linear displacement coefficients; The reflection area coordinates are translated along the X-axis by Δx and along the Y-axis by Δy to generate the dynamic reflection area prediction coordinates; The calibration process of the linear displacement coefficient is as follows: In the initialization stage, the user is guided to perform a standard posture action; Based on the ratio of the actual reflection area displacement amount Δx' and Δy' to the measured angle β' and α', k1=Δx' / β' and k2=Δy' / α' are updated. 4.The AI-based indoor lighting simulation and optimization method of claim 1, wherein: The method comprises the following steps: The line of sight focus point coordinates are expanded to a preset radius circular area with the focus point as the center; The geometric overlap area ratio of the preset radius circular area and the dynamic reflection area prediction coordinates is calculated as the overlap rate; When the overlap rate exceeds the first-level optimization threshold and the average brightness of the reflection area is >150, a first-level optimization instruction is generated; When the overlap rate is between the second-level optimization threshold and the first-level optimization threshold, a second-level optimization instruction is generated; The first-level optimization threshold and the second-level optimization threshold are linearly reduced according to the user fatigue data, and the user fatigue data is the continuous staring time of the user at the screen. 5.The AI-based indoor lighting simulation and optimization method of claim 4, wherein: The method comprises the following steps: Calculate the brightness contribution of each luminaire to the light-reflecting area: Wherein, C i represents the brightness contribution of the i-th luminaire to the light-reflecting area, L i is the luminaire brightness, θ i is the angle between the luminaire light and the normal of the light-reflecting area, d i is the distance between the luminaire and the light-reflecting area; normalizing the luminance contribution of each luminaire to the reflective region to a weight of each luminaire In response to the first-level optimization instruction, the lamp with the highest weight is adjusted, and its brightness is reduced to: L adjusted = L i • (1 - W i • 0.5); The tilt angle adjustment amount is: In response to the second-level optimization instruction, the color temperature of the lamps with the top two weights is synchronously adjusted to: CCT adjusted = CCT i - (W i 500), and no less than 2500 K. 6.The AI-based indoor lighting simulation and optimization method of claim 5, wherein: When the user's continuous gaze on the same area is detected for more than a preset time threshold, the weight of the light associated with the area is increased by 50%; When the user's head rotation angle is detected to be more than a preset angle threshold, the weight of the light associated with the new gaze area is reset to the initial value. 7.The AI-based indoor lighting simulation and optimization method of claim 1, wherein: The process of adjusting the linear displacement coefficient in proportion is: Obtain the brightness drop amplitude ΔL; If ΔL≥15%, then: And when ΔL<15%, mark the current lamp parameter combination as an inefficient parameter and reduce its historical weight to 0.

2. 8.The AI-based indoor lighting simulation and optimization method of claim 1, wherein: When the brightness drop amplitude is less than 20%, trigger the physical light shield control instruction; when the brightness drop amplitude is ≥20%, store the current skeletal key point coordinates, displacement amount, and lamp parameter combination to the historical optimization database. 9.The AI-based indoor lighting simulation and optimization method of claim 1, wherein: After generating the optimization instruction, it further includes: Based on the lamp parameter adjustment records of the same skeletal key point coordinates in the historical optimization database, preferentially select the parameter combinations verified to be effective in history to perform adjustment; The process of preferentially selecting the parameter combinations verified to be effective in history to perform adjustment includes: Statistically combine historical success rates of parameter combinations Generate priority weight W priority ; When the same bone key point coordinates are newly detected, the W is preferentially called priority The parameter combination of > 0.

8.

10. An AI-based indoor lighting simulation and optimization system, characterized by: Use the AI-based indoor lighting simulation and optimization method as claimed in any one of claims 1-9, including: A data acquisition module for acquiring user's skeletal key point coordinates, screen reflection area coordinates, and line of sight focus coordinates; A dynamic prediction module for calculating the displacement amount of the reflection area based on the offset angle of the skeletal key point coordinates through a preset linear coefficient to generate dynamic reflection area prediction coordinates; An overlap determination module for calculating the overlap rate of the line of sight focus coordinates and the dynamic reflection area prediction coordinates, and generating an optimization instruction when the overlap rate exceeds a preset overlap rate threshold; A light source control module for adjusting the brightness and tilt angle of the preset associated light according to the optimization instruction, so that the adjusted actual reflection area brightness drops below a preset safe brightness threshold; A verification feedback module for re-acquiring the actual reflection area brightness value after adjusting the lamp parameters, calculating the brightness drop amplitude, and adjusting the linear displacement coefficient in proportion according to the difference between the brightness drop amplitude and the preset target value; A historical data management module for statistically analyzing the historical success rate of each parameter combination in the historical optimization database, and preferentially selecting the parameter combinations verified to be effective in history to perform adjustment based on the lamp parameter adjustment records of the same skeletal key point coordinates in the historical optimization database.

Images