Cursor control system based on monochromatic stick image recognition

By using monochrome stick image recognition technology, combined with adaptive lighting compensation and machine learning optimization, the accuracy and stability issues of existing cursor control technologies in large screens and complex lighting environments have been resolved, achieving efficient and accurate cursor control and improving the user experience.

WO2026233242A1PCT designated stage Publication Date: 2026-11-12CHEN BIHONG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2026/094315
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-06
Filing Date
2026-04-30
Publication Date
2026-11-12

AI Technical Summary

Technical Problem

Existing cursor control technology suffers from insufficient precision in large-screen displays and complex lighting environments, poor environmental adaptability, and inadequate real-time performance and stability, making it difficult to achieve efficient and accurate cursor control.

Method used

A cursor control system based on monochrome stick image recognition is adopted. The system acquires video frames through a camera and converts them to the HSV color space. It extracts monochrome stick-shaped objects by combining Canny edge detection and contour analysis algorithms. It uses adaptive illumination compensation and three-dimensional space compensation algorithms to perform accurate coordinate mapping. The action recognition module determines the operation command through displacement vector and angular velocity, and optimizes the detection parameters through machine learning to achieve multi-user interaction and anti-occlusion processing.

Benefits of technology

It achieves high-precision cursor control in complex environments, improves control stability and real-time response capabilities, adapts to different lighting conditions, supports multi-user interaction, and significantly enhances the user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2026094315_12112026_PF_FP_ABST
    Figure CN2026094315_12112026_PF_FP_ABST
Patent Text Reader

Abstract

Disclosed in the present invention is a cursor control system based on monochromatic stick image recognition, the system comprising: an image acquisition module, which is configured to acquire in real time consecutive video frames including a monochromatic stick-shaped object by means of a camera; a preprocessing module, which is configured to convert the video frames from an RGB color space to an HSV color space and establish a monochromatic hue threshold range model; an object detection module, which is configured to perform noise reduction processing on a binarized image by means of a morphological operation and use Canny edge detection combined with a contour analysis algorithm to extract a monochromatic stick-shaped object that satisfies the following geometric features; a coordinate mapping module, which is configured to convert geometric center coordinates of the detected monochromatic stick-shaped object into screen cursor coordinates by means of a bilinear interpolation algorithm; an action recognition module, which is configured to determine an operation instruction by calculating a displacement vector magnitude and an angular velocity of the monochromatic stick-shaped object between consecutive frames; and a control module, which is configured to map a geometric center position of the monochromatic stick-shaped object to a screen coordinate system and execute a click operation in response to a single-click signal.
Need to check novelty before this filing date? Find Prior Art

Description

A cursor control system based on monochrome stick image recognition Technical Field

[0001] This invention relates to the field of human-computer interaction technology, specifically to a cursor control system based on monochrome stick image recognition. Background Technology

[0002] With the advancement of technology, human-computer interaction methods are constantly being innovated, from traditional keyboards and mice to touch screens, and then to gesture recognition-based interaction methods, resulting in a significant improvement in user experience.

[0003] Existing cursor manipulation technologies mainly include the following:

[0004] Traditional input devices, such as mice, touchpads, and keyboards, offer high precision and responsiveness in desktop environments. However, in applications such as large-screen displays and smart TVs, the user experience with traditional input devices is poor, requiring users to frequently move their arms or stand up to operate, leading to fatigue and inconvenience.

[0005] Touchscreen technology: Touchscreen technology is widely used in devices such as smartphones and tablets, allowing users to operate the device by directly touching the screen, providing an intuitive interaction method. However, the use of touchscreens is limited on large-screen devices, requiring users to be close to the screen to operate it, and operational conflicts can occur when multiple people share the screen.

[0006] Camera-based 2D gesture recognition technology: This technology captures user gestures using a camera to enable contactless control. For example, devices like Kinect and LeapMotion utilize depth sensors and computer vision algorithms to recognize user gestures and convert them into control commands. However, these technologies suffer from low accuracy in complex lighting conditions and have limited recognition range and precision. Furthermore, existing technologies largely rely on complex machine learning algorithms, and their real-time performance and stability need improvement.

[0007] Infrared sensor-based control technology: Some smart TVs and game consoles use infrared sensors to capture user movements and achieve simple control functions. However, these technologies have limited recognition accuracy and response speed, making it difficult to achieve precise cursor control, and are easily affected by ambient light.

[0008] Despite the progress made in the field of human-computer interaction, the following technical problems still exist in practical applications:

[0009] Insufficient control precision: Traditional input devices and touch screens cannot provide sufficient control precision in certain application scenarios, especially when displaying on large screens and operating from a distance, resulting in a poor user experience.

[0010] Poor environmental adaptability: Camera-based 2D gesture recognition technology and infrared sensor-based control technology have low recognition accuracy in complex lighting environments, are easily affected by ambient light, resulting in unstable control, and have limited recognition range and accuracy for user gestures.

[0011] Insufficient real-time performance and stability: Existing technologies largely rely on complex machine learning algorithms, and their real-time performance and stability need improvement. During rapid movement or continuous operation, the recognition and response speeds may not keep up, impacting the user experience.

[0012] Therefore, a cursor control system based on monochrome stick image recognition is proposed to address the current shortcomings. Summary of the Invention

[0013] The technical problem to be solved by the present invention is to overcome the defects of the above-mentioned technology and provide a cursor control system based on monochrome stick image recognition.

[0014] To solve the above-mentioned technical problems, the technical solution provided by the present invention is a cursor control system based on monochrome stick image recognition: including...

[0015] The image acquisition module is used to acquire continuous video frames containing monochrome rod-shaped objects in real time via a camera;

[0016] The preprocessing module converts the video frames from the RGB color space to the HSV color space and establishes a monochrome hue threshold range model: H min ∈[35,77] S min ≥50 V min ≥50

[0017] Where: H represents hue value, S represents saturation, and V represents lightness value;

[0018] The object detection module performs noise reduction on the binarized image through morphological operations. It employs a Canny edge detection algorithm combined with contour analysis to extract monochrome rod-shaped objects that meet the following geometric characteristics: aspect ratio ≥ 3:1, contour area ≥ 100 pixels, and circularity ≤ 0.3.

[0019] The coordinate mapping module maps the geometric center coordinates (x, y) of the detected monochromatic rod-shaped object. c y c Convert the coordinates to screen cursor coordinates (X, Y) using bilinear interpolation:

[0020] Among them, (x min y min ) and (x max y max W represents the boundary coordinates of the image acquisition area. screenand H screen For screen display resolution; motion recognition module, which calculates the displacement vector magnitude of a monochrome rod-shaped object between consecutive frames. Operation commands for determining angular velocity ω:

[0021] When satisfied A click event is triggered when the pixel angle (ω) changes by more than 45°.

[0022] Control module: Maps the geometric center position of the monochrome rod to the screen coordinate system and executes the click operation in response to the click signal.

[0023] As an improvement, the preprocessing module includes an adaptive illumination compensation unit, which dynamically adjusts the V channel using a histogram equalization algorithm. The compensation coefficient is calculated using the following formula:

[0024] Among them, V target The target brightness value is set between 180 and 220.

[0025] As an improvement, the coordinate mapping module integrates a three-dimensional space compensation algorithm, and introduces a perspective transformation matrix when the camera and the display screen are not orthogonal.

[0026] The matrix parameters are obtained in advance using the Zhang Zhengyou calibration method.

[0027] As an improvement, an anti-occlusion unit is also included. When the monochrome stick is briefly occluded, the cursor position continuity is maintained based on the motion prediction model. The prediction formula is:

[0028] Among them, v x v y The instantaneous velocity of the geometric center of the monochromatic rod before it is blocked.

[0029] As an improvement, the action recognition module includes a multi-user interaction unit, which allocates control rights according to the following priority rules when multiple monochrome stick-shaped objects are detected:

[0030] The target with the largest area has the highest priority;

[0031] Newly entered targets gain temporary priority (lasting 3 seconds);

[0032] When targets overlap by more than 50%, a competition mode is activated, and the target with the faster displacement speed gains control.

[0033] As an improvement, an environment adaptation module is also included, which continuously optimizes the detection parameters through machine learning algorithms. This algorithm employs an improved YOLOv5 network structure, and the loss function is defined as: L = λcls L cls +λ obj L obj +λ box L box

[0034] Where, λ cls =0.5, λ obj =1.0, λ box =1.5.

[0035] As an improvement, the camera is a wide-angle lens with a field of view θ≥120° and a frame rate f≥30fps, which is suitable for cursor control in large space scenarios.

[0036] As an improvement, a reflective marker or infrared emitter can be attached to the end of the monochrome stick to achieve multimodal positioning in conjunction with a camera, thereby enhancing the robustness of recognition under complex lighting conditions.

[0037] As an improvement, the control module supports protocol compatibility with external devices to enable cross-platform control of televisions, computers, or smart terminals.

[0038] As an improvement, the protocols of the external device include HDMI-CEC, USB HID, and Bluetooth interfaces.

[0039] The advantages of this invention compared to the prior art are:

[0040] 1. This invention achieves high-precision cursor control through precise geometric feature extraction and coordinate mapping algorithms. Utilizing Canny edge detection and contour analysis algorithms, it can accurately identify the geometric center of a monochrome rod-shaped object and map it to the screen coordinate system, thereby achieving precise cursor positioning.

[0041] 2. This invention, through HSV color space conversion and an adaptive illumination compensation unit, can dynamically adjust the brightness value of an image to compensate for the effects of changes in illumination. Furthermore, the improved YOLOv5 network structure further enhances the robustness of recognition under complex backgrounds and varying illumination conditions, ensuring stable operation of the system in different environments.

[0042] 3. This invention, through online learning and dynamic weight adjustment mechanisms, can continuously optimize detection parameters based on real-time detection data, thereby improving the system's real-time response capability. Simultaneously, when a monochrome stick is briefly occluded, the motion prediction model can predict its position based on historical velocity, maintaining the continuity of the cursor position and thus improving the overall stability of the system.

[0043] 4. This invention, through a multi-user interaction unit and priority rules, can effectively distinguish the handheld monochrome stick-shaped objects of different users, and allocate control rights according to parameters such as area, order of entry into the field of vision, and displacement speed, so as to ensure stable operation in a multi-user environment.

[0044] 5. This invention, through an anti-occlusion unit based on a motion prediction model, can predict the position of a monochrome stick-shaped object in the next frame when it is occluded, maintaining the continuity of the cursor position and improving the user's control experience.

[0045] 6. This invention uses an environment adaptive module and machine learning algorithms to continuously optimize detection parameters, which can automatically adjust system parameters according to different environmental conditions, thereby improving the system's adaptability and detection accuracy in different environments.

[0046] 7. Based on a solid color bar, this invention achieves efficient, precise and user-friendly cursor control through the operation of the solid color bar, significantly improving the human-computer interaction experience. Attached Figure Description

[0047] Figure 1 is a schematic diagram of the structure of a cursor control system based on monochrome stick image recognition according to the present invention.

[0048] [Correction 19.05.2026 based on Rule 91] Figure 2 is a system diagram of a cursor control system based on monochrome stick image recognition according to the present invention.

[0049] [Corrected on May 19, 2026, according to Rule 91]

[0050] [Corrected on May 19, 2026, according to Rule 91] Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the invention clearer, the technical solutions of the embodiments of the invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the invention, not all embodiments. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0052] In the description of the embodiments of the invention, it should be noted that if terms such as "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," or "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on the invention. Furthermore, terms such as "first," "second," and "third" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0053] Furthermore, the use of terms such as "horizontal," "vertical," and "sag" does not imply that the component must be absolutely horizontal or suspended, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal relative to "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0054] In the description of the embodiments of the invention, "a plurality of" means at least two.

[0055] In the description of the embodiments of the invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the invention according to the specific circumstances.

[0056] As shown in the accompanying drawings, a cursor control system based on monochrome stick image recognition includes...

[0057] The image acquisition module is used to acquire continuous video frames containing monochrome rod-shaped objects in real time through a camera, providing necessary input data for subsequent processing modules.

[0058] Specifically, the camera is a wide-angle lens, suitable for cursor control in large space scenes; in this embodiment, the field of view θ≥120°, covering the monochrome stick movement range frame rate in large space scenes; the frame rate f≥30fps, ensuring the capture of smooth motion trajectories and avoiding ghosting or delay.

[0059] In practice, reflective markers or infrared emitters can be attached to the end of the monochrome stick to achieve multimodal positioning in conjunction with a camera, thereby improving the robustness of recognition in complex lighting environments.

[0060] The preprocessing module converts the video frames from the RGB color space to the HSV color space and establishes a monochrome hue threshold range model: H min ∈[35,77] S min ≥50 V min ≥50

[0061] Where: H represents hue value, S represents saturation, and V represents lightness value;

[0062] In practice, if a pixel has an RGB value of (30, 200, 80) and a converted HSV value of (50, 85, 90), it meets the threshold condition and is retained as a candidate region.

[0063] Furthermore, the preprocessing module includes an adaptive illumination compensation unit, which dynamically adjusts the V channel using a histogram equalization algorithm. The compensation coefficient is calculated using the following formula:

[0064] Among them, V target The target brightness value is set between 180 and 220.

[0065] During implementation:

[0066] 1) Target brightness value: V target =200 (take the middle value).

[0067] 2) Calculation of average lightness:

[0068] Assuming the average brightness of the current frame is V ang =150, satisfying V ang ≥128.

[0069] 3) Calculation of compensation coefficient

[0070] 4) V-channel adjustment:

[0071] For each pixel, the new brightness value V new =V old ×k comp

[0072] Example pixel adjustment: V new =V old ×k comp =100×1.333≈133 (satisfies V) min >50)

[0073] Therefore, by increasing the brightness of low-light areas (e.g., increasing V from 100 to 133), the contrast between the monochrome stick and the background is enhanced.

[0074] In practice,

[0075] HSV threshold segmentation:

[0076] 4.1 Specific Calculation Formula

[0077] 4.2 Example Data:

[0078] The original HSV value of the pixel is (50, 85, 90);

[0079] After dynamic illumination compensation, the HSV value is (H=50, S=85, V=120).

[0080] judge:

[0081] H=50∈[35,77],S=85≥50,

[0082] That is, select pixels that meet the monochrome characteristics (such as monochrome sticks) and exclude other color interference (such as red and blue objects).

[0083] The object detection module performs noise reduction on the binarized image through morphological operations. It employs a Canny edge detection algorithm combined with contour analysis to extract monochrome rod-shaped objects that meet the following geometric characteristics: aspect ratio ≥ 3:1, contour area ≥ 100 pixels, and circularity ≤ 0.3.

[0084] In practice:

[0085] Candidate contour filtering:

[0086] 1) Aspect Ratio:

[0087] Example: If the outline width is 150 pixels and the height is 50 pixels → the aspect ratio is 3:1, which meets the requirements.

[0088] 2) Outline area: Area = Number of pixels ≥ 100

[0089] Example: The outline contains 200 pixels → meets the requirements.

[0090] 3) Circularity

[0091] Example: If the perimeter is 100 pixels and the area is 200 pixels, the circularity is approximately 0.628 (not compliant, rejected).

[0092] The coordinate mapping module maps the geometric center coordinates (x, y) of the detected monochromatic rod-shaped object. c y c Convert the coordinates to screen cursor coordinates (X, Y) using bilinear interpolation:

[0093] Among them, (x min ymin ) and (x max y max W represents the boundary coordinates of the image acquisition area. screen and H screen For screen display resolution;

[0094] Furthermore, the coordinate mapping module integrates a three-dimensional space compensation algorithm, introducing a perspective transformation matrix when the camera and display screen are not orthogonal.

[0095] The matrix parameters are obtained in advance using the Zhang Zhengyou calibration method.

[0096] In practical implementation:

[0097] 1. Input data

[0098] 1.1 Geometric center coordinates of the monochromatic rod:

[0099] The coordinates of the center of the monochrome rod-shaped object obtained by the target detection module are (x... c y c = (960, 500) (pixels, based on the image acquisition area).

[0100] 2. Image acquisition area boundary:

[0101] Top left corner coordinates: (x min y min ) = (100, 50);

[0102] bottom right corner coordinates: (x max y max ) = (1820, 930);

[0103] 3. Screen resolution:

[0104] Width: W screen =1920 pixels;

[0105] Height: H screen =1080 pixels;

[0106] 4. Perspective transformation matrix (calibrated using Zhang Zhengyou's calibration method)

[0107] 5. Coordinate mapping calculation:

[0108] 5.1 Lateral (X-axis) Mapping

[0109] That is, the monochrome stick-shaped object is located in the center of the image and remains in the center of the screen after mapping.

[0110] 5.2 Vertical (Y-axis) Mapping

[0111] 5.3 Mapping Results:

[0112] The screen coordinates are (X, Y) = (960, 607). At this time, the monochrome rod-shaped object is located slightly above the center of the screen.

[0113] When implementing this, if the camera and screen are not orthogonal (e.g., the camera is looking down while the screen is placed horizontally), perspective distortion needs to be corrected.

[0114] At this point, the perspective transformation formula is:

[0115] Specifically:

[0116] Image coordinates (x) c y c 1) Substitute into the matrix:

[0117] That is: X = 900.8Y = 548

[0118] Final screen coordinates:

[0119] (X, Y) = (901, 548), which corrects the coordinate offset caused by the camera looking down.

[0120] It also includes an anti-occlusion unit, which maintains the continuity of the cursor position based on a motion prediction model when the monochrome stick is briefly occluded. The prediction formula is:

[0121] Among them, v x v y The instantaneous velocity of the geometric center of the monochromatic rod before it is blocked.

[0122] In practice, when a monochrome rod-shaped object is briefly occluded, its position in the next frame is predicted based on its historical velocity.

[0123] 1. Coordinates of the last frame before the monochrome rod-shaped object was occluded: (x c y c ) = (960, 500);

[0124] 2. Speed ​​of the monochrome roller before obstruction:

[0125] v x = 3 pixels / frame (horizontally to the right);

[0126] v y = -2 pixels / frame (vertically downward);

[0127] 3. Time interval: Δt = 1 frame;

[0128] 4. Predict the coordinates of the next frame:

[0129] 4.1 Horizontal Forecasting

[0130] 4.2 Vertical Forecasting

[0131] Predict the coordinates of the next frame: x^c = 960 + 3 × 1 = 963 y^c = 500 - 2 × 1 = 498.

[0132] 5. Predicted cursor position during occlusion

[0133] 5.1 Image Coordinates

[0134] 5.2 Screen coordinates (converted via coordinate mapping module): X = 1084 pixels, Y = 523 pixels

[0135] 6. Visualization of prediction results:

[0136] 6.1 Before occlusion: The monochrome stick is located in the center of the screen (X=960, Y=607);

[0137] 6.2 During occlusion: The cursor is predicted to be X=1084, Y=523 based on historical speed (located in the upper right corner of the screen).

[0138] 6.3 After visibility is restored: If the monochrome stick reappears, the system immediately switches back to the actual detection coordinates; if it does not reappear, the prediction continues until the occlusion timeout (e.g., 3 seconds).

[0139] The motion recognition module calculates the displacement vector magnitude of a monochromatic stick-shaped object between consecutive frames. Operation commands for determining angular velocity ω:

[0140] When satisfied A click event is triggered when the pixel angle (ω) changes by more than 45°.

[0141] In practice,

[0142] 1. Input data

[0143] The center coordinates of the monochrome stick in consecutive frames:

[0144] Frame t: (x c,t y c,t ) = (960, 500)

[0145] Frame t-1: (x c,t-1 y c,t-1 ) = (945, 505)

[0146] 2. Calculation of displacement vector magnitude

[0147] The displacement modulus is 15.81 pixels, which is greater than 15 pixels, satisfying the displacement threshold for a click event.

[0148] 3. Angular velocity calculation

[0149] An angular velocity of -18.43° indicates that the monochromatic stick is moving to the lower right with a gradual change in direction.

[0150] 3. Results Analysis:

[0151] Amplitude of angular velocity change | ω-prev ω |=|-18.43°-0°|=18.43°<45°

[0152] The displacement modulus is 15.81 pixels, which is greater than 15, but the angular velocity change amplitude is less than 45°, which does not meet the requirement of triggering a click event.

[0153] Furthermore, the action recognition module includes a multi-user interaction unit, which allocates control rights according to the following priority rules when multiple monochrome stick-shaped objects are detected:

[0154] The target with the largest area has the highest priority;

[0155] Newly entered targets gain temporary priority (lasting 3 seconds);

[0156] When targets overlap by more than 50%, a competition mode is activated, and the target with the faster displacement speed gains control.

[0157] During implementation, when two monochromatic rollers (A and B) are detected simultaneously, the parameters are as follows:

[0158] Monochrome rod-shaped object A:

[0159] Area: 300 pixels;

[0160] Displacement modulus: 20 pixels;

[0161] Angular velocity: 30°;

[0162] Monochrome rod-shaped object B:

[0163] Area: 150 pixels;

[0164] Displacement modulus: 10 pixels;

[0165] Angular velocity: 5°;

[0166] Priority determination:

[0167] Area priority: Monochrome rod A has a larger area (300>150) and receives the highest priority.

[0168] New targets temporarily prioritized: If the monochrome stick-shaped object B is newly entered into the field of view and its duration is less than 3 seconds, its priority is temporarily increased.

[0169] Competition mode: If the overlap area between the two is >50%, compare the speed.

[0170] Ultimate control: The single-color stick A triggers a click event.

[0171] It also includes an environment adaptation module that continuously optimizes detection parameters through machine learning algorithms. These algorithms employ an improved YOLOv5 network structure, and the loss function is defined as: L = λ cls L cls +λ obj L obj +λ box L box

[0172] Where, λ cls =0.5, λ obj =1.0, λ box =1.5.

[0173] During implementation:

[0174] 1. Input data

[0175] 1.1 Training Dataset:

[0176] Includes 5000 labeled images, covering the following scenarios:

[0177] Lighting conditions: 500-2000 lux (simulating indoor / outdoor light variations);

[0178] Background complexity: solid color background, textured background (such as wood grain, fabric), dynamic background (such as moving people);

[0179] Target state: The monochrome stick is intact or partially obscured (e.g., covered by a finger);

[0180] 1.2 Real-time detection data:

[0181] Continuous video frames from a camera (1920×1080 resolution, 30fps), including monochrome stick motion and environmental changes.

[0182] 2. Loss Calculation:

[0183] Suppose that in a certain batch of training:

[0184] Classification loss L cls =0.3;

[0185] Target loss L obj =0.5;

[0186] Bounding box loss L box =0.8;

[0187] The total loss is: L = 0.5 × 0.3 + 1.0 × 0.5 + 1.5 × 0.8 = 0.15 + 0.5 + 1.2 = 1.85

[0188] 3. Dynamic optimization

[0189] 3.1 Online learning

[0190] Triggering conditions:

[0191] If the total loss L > 1.5 after every 100 frames detected, online fine-tuning is initiated.

[0192] Optimization process:

[0193] a. Freeze the backbone network: Keep the backbone feature extraction layer of YOLOv5s unchanged, and only train the newly added infrared detector head and the last fully connected layer.

[0194] b. Gradient descent: Update network weights using the Adam optimizer with a learning rate of 10⁻⁴.

[0195] c. Stopping condition: Training is terminated when the total loss of 5 consecutive batches decreases by less than 0.1%.

[0196] 3.2 Dynamic weight adjustment:

[0197] 3.2.1 Adaptive to changes in illumination:

[0198] Based on the current frame average brightness V avg Dynamically adjust classification loss weights:

[0199] but:

[0200] Daytime (V) avg =200):λ cls =0.8 (enhancing classification ability).

[0201] Night (V) avg =100):λ cls =0.5 (Balanced classification and positioning).

[0202] 3.2.2 Optimization of Occlusion Scenes:

[0203] If detection fails for 3 consecutive frames, increase the confidence loss weight: λ obj = 1.0 + 0.5 * occlusion duration (s)

[0204] Control module: Maps the geometric center position of the monochrome rod to the screen coordinate system and executes the click operation in response to the click signal.

[0205] Specifically, the control module supports protocol compatibility with external devices to enable cross-platform control of televisions, computers, or smart terminals.

[0206] In practice, the protocols of the external device include HDMI-CEC, USB HID, and Bluetooth interfaces.

[0207] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A cursor control system based on monochrome stick image recognition, characterized in that, include The image acquisition module is used to acquire continuous video frames containing monochrome rod-shaped objects in real time via a camera; The preprocessing module converts the video frames from the RGB color space to the HSV color space and establishes a monochrome hue threshold range model. H min ∈[35,77] S min ≥50 V min ≥50 Where: H represents hue value, S represents saturation, and V represents lightness value; The object detection module performs noise reduction on the binarized image through morphological operations and uses the Canny edge detection combined with contour analysis algorithm to extract monochrome stick-shaped objects that meet the following geometric features: Aspect ratio ≥ 3:1; outline area ≥ 100 pixels; roundness ≤ 0.3 The coordinate mapping module maps the geometric center coordinates (x, y) of the detected monochromatic rod-shaped object. c y c Convert the coordinates to screen cursor coordinates (X, Y) using bilinear interpolation: Among them, (x min y min ) and (x max y max W represents the boundary coordinates of the image acquisition area. screen and H screen For screen display resolution; The motion recognition module calculates the displacement vector magnitude of a monochromatic stick-shaped object between consecutive frames. Operation commands for determining angular velocity ω: When satisfied A click event is triggered when the pixel angle (ω) changes by more than 45°. Control module: Maps the geometric center position of the monochrome rod to the screen coordinate system and executes the click operation in response to the click signal.

2. The cursor control system based on monochrome stick image recognition according to claim 1, characterized in that: The preprocessing module includes an adaptive illumination compensation unit, which dynamically adjusts the V channel using a histogram equalization algorithm. The compensation coefficient is calculated using the following formula: Among them, V target The target brightness value is set between 180 and 220.

3. The cursor control system based on monochrome stick image recognition according to claim 1, characterized in that: The coordinate mapping module integrates a three-dimensional space compensation algorithm. When the camera and the display screen are not orthogonal, a perspective transformation matrix is ​​introduced. The matrix parameters are obtained in advance using the Zhang Zhengyou calibration method.

4. A cursor control system based on monochrome stick image recognition according to claim 1, characterized in that: It also includes an anti-occlusion unit, which maintains the continuity of the cursor position based on the motion prediction model when the monochrome stick is briefly occluded. The prediction formula is: Among them, v x v y The instantaneous velocity of the geometric center of the monochromatic rod before it is blocked.

5. A cursor control system based on monochrome stick image recognition according to claim 1, characterized in that: The action recognition module includes a multi-user interaction unit. When multiple monochrome stick-shaped objects are detected, control rights are allocated according to the following priority rules: The target with the largest area has the highest priority; Newly entered targets gain temporary priority (lasting 3 seconds); When targets overlap by more than 50%, a competition mode is activated, and the target with the faster displacement speed gains control.

6. A cursor control system based on monochrome stick image recognition according to claim 1, characterized in that: It also includes an environment adaptation module that continuously optimizes detection parameters through machine learning algorithms. These algorithms employ an improved YOLOv5 network structure, and the loss function is defined as: L = λ cls L cls +λ obj L obj +λ box L box Among them, l cls =0.5,λ obj =1.0,λ box = 1.

5.

7. A cursor control system based on monochrome stick image recognition according to claim 1, characterized in that: The camera is a wide-angle lens with a field of view θ≥120° and a frame rate f≥30fps, suitable for cursor control in large space scenarios.

8. A cursor control system based on monochrome stick image recognition according to claim 1, characterized in that: The end of the monochrome stick can be attached with a reflective marker or an infrared emitter to work with a camera to achieve multimodal positioning and improve the robustness of recognition in complex lighting environments.

9. A cursor control system based on monochrome stick image recognition according to claim 1, characterized in that: The control module supports protocol compatibility with external devices to enable cross-platform control of televisions, computers, or smart terminals.

10. A cursor control system based on monochrome stick image recognition according to claim 9, characterized in that: The protocols of the external devices include HDMI-CEC, USB HID, and Bluetooth interfaces.