Fan adaptive following steering method based on ai decision

Abstract

The application belongs to the technical field of intelligent fan steering control, and particularly discloses a fan adaptive following steering method based on AI decision, which comprises the following steps: step one, obtaining current device setting information of the fan, encoding the device setting information into a device state feature vector, taking the device state feature vector as a subsequent initial reference representation, and uniformly integrating into a joint vector space. The fan adaptive following steering method based on AI decision is based on human trace positioning in the application, blows air only in the space range where people are present, automatically and quickly skips the area where no one is present, so that the user who needs to be blown by the air can obtain a longer effective air supply time in a unit time, air blowing is automatically stopped when no one is present for more than a set time through sensor monitoring, and air blowing is immediately responded to when a person appears, so that energy waste is eliminated.

Description

Technical Field

[0001] This invention relates to the field of intelligent fan steering control technology, specifically to a fan adaptive following steering method based on AI decision-making. Background Technology

[0002] Fans are common cooling appliances. With the development of smart home technology, the level of intelligence of fans has also improved. Traditional fans mostly rely on manual adjustment or simple timer and oscillation functions. In recent years, smart fans have turned on when a person is detected by a single pyroelectric sensor.

[0003] However, the machine is running while people are present, so it cannot accurately blow air onto the people who need it. It also blows air into idle areas, wasting energy and causing too much time between people who need it. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based fan adaptive steering method to solve the problem of energy waste caused by single-sensor air supply in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a fan adaptive following steering method based on AI decision-making, comprising the following steps: Step 1: Obtain the current device settings information of the fan, encode the device settings information into a device status feature vector, use it as the initial benchmark representation for subsequent AI decision-making, and uniformly incorporate it into the joint vector space; Step 2: Collect data within a set range in real time using sensors, map each feature value to a unified high-dimensional vector space, and generate a direction information vector; Step 3: Input the direction information and the current fan status feature value into the AI ​​decision unit to generate a sequence of action commands; Step 4: Based on the continuous analysis results of the AI ​​decision-making unit, realize the intelligent steering of the fan; Preferably, the sensors in step two include, but are not limited to, cameras, depth sensors, microphone arrays, and infrared pyroelectric detectors, and the data in step two include visual images, depth maps, sound source locations, and thermal distribution.

[0006] By adopting the above technical solution, robustness in complex environments can be improved and target loss can be avoided due to visual occlusion.

[0007] Preferably, the direction vector in step two includes human position, movement trajectory, identity features and acoustic and thermal cues. In step two, the spatiotemporal correlation and mutual information are maintained through a collaborative representation mechanism, which improves the robustness of perception in complex scenarios such as occlusion and low light.

[0008] By adopting the above technical solution, the robustness of system perception can be increased through the collaborative representation mechanism.

[0009] Preferably, in step three, the AI ​​decision-making unit has built-in cross-task understanding and planning capabilities. These capabilities combine the scenario to perform intent reasoning, determine the type of crowd behavior, and combine it with user preset preferences. Through an attention mechanism, weights are dynamically allocated to generate the optimal sequence of action instructions. The type of crowd behavior includes whether the crowd is passing by briefly or staying to cool off. The user preset preferences include single-person locking, timed settings, and energy saving. The sequence of action instructions includes the fan's horizontal turning angle, the fan's vertical turning angle, the sweeping speed, and the duration of stay in the area.

[0010] By adopting the above technical solution, precise air delivery can be achieved, improving the user experience.

[0011] Preferably, the fan modes in step four include single-target continuous tracking mode, multi-person on-demand coverage mode, and intelligent servo standby mode.

[0012] By adopting the above technical solution, different usage scenarios can be achieved by switching between multiple fan modes.

[0013] Preferably, the specific content of the single-target continuous tracking mode is as follows: set a specific target to be tracked, lock the feature ID of the target in the joint vector space, and perform "sensing-decision-control" closed-loop tracking to realize the following of the moving target.

[0014] Using the above technical solution, the following of a moving target can be achieved through a single-target continuous tracking mode.

[0015] Preferably, the multi-person on-demand coverage mode specifically involves: identifying and marking human body distribution hot zones, with the fan only delivering air to areas where effective human characteristics are detected in the joint characterization space, quickly skipping unoccupied areas, and maximizing the energy efficiency of airflow in areas with people.

[0016] By adopting the above technical solution, the multi-person on-demand coverage mode can achieve the goal of blowing only in the area where people are present.

[0017] Preferably, the intelligent servo standby mode specifically includes the following: when no human feature vector is matched in the joint representation space after a preset time, the system determines that there is no one in the environment, and the fan autonomously enters a low-power servo state; the sensor continues to collect data, and when human activity features reappear in the joint representation space, it immediately triggers wake-up and restores the air supply state according to the original settings.

[0018] Using the above technical solution, the intelligent servo standby mode can automatically supply air and stop supplying air.

[0019] Preferably, the method uses a local chip to implement the algorithm. The chip is connected to the target fan and the sensor. The chip has a built-in target detection algorithm that detects all people in a single frame image and marks their positions with bounding boxes. The chip monitors the target situation within its real-time range based on pixel coordinates and camera intrinsic parameters. The detected center of the human body in the image is converted into a direction vector emanating from the optical center of the camera, thus obtaining the horizontal rotation angle. Given the camera's horizontal field of view (HFOV), vertical field of view (VFOV), image width (W), and height (H), the formula for calculating the horizontal rotation angle is:

[0020] Given the camera's horizontal field of view (HFOV), vertical field of view (VFOV), image width (W), and height (H), the formula for calculating the vertical rotation angle is: .

[0021] The above technical solution can provide a control basis for fan following.

[0022] Preferably, the method further includes directional air supply control, the specific process of which is as follows: Visual computing: Detecting the center of the human body and calculating the generated orientation angle. ; Angle mapping: mapping the optical axis direction angle of the camera Converted to the absolute target angle of the fan mechanical shaft The rotation matrix is ​​calibrated based on the camera's installation offset. Control quantity generation: For servos, calculate pulse width or pulse number and direction signal; for DC motors, input parameters into the controller to output PWM duty cycle. Control quantity generation: The drive motor rotates using a closed-loop system, and the actual angle is fed back by sensors to form a control closed loop.

[0023] The above technical solution can achieve directional air supply.

[0024] Compared with existing technologies, the beneficial effects of this invention are: the AI-based fan adaptive following steering method: 1. Based on human location, this invention delivers air only to areas where people are present, automatically and quickly skipping unoccupied areas, allowing users who need airflow to obtain a longer effective airflow duration per unit time; using sensors to monitor, the airflow automatically stops when no one is present for a set time, and responds immediately when someone appears, eliminating energy waste. 2. In this invention, redundant modalities such as sound sources can still be fused even under visual obstruction or changes in light to maintain continuous tracking of moving targets and avoid losing track of them. In multi-person scenarios, the AI ​​decision-making unit can extract and lock a unique target according to user settings, and only track and deliver air to that target, while other personnel are not disturbed. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the process structure of the present invention; Figure 2 This is a schematic diagram of the multi-person airspace range structure of the present invention; Figure 3 This is a schematic diagram of the fan single-target tracking structure of the present invention; Figure 4 This is a schematic diagram of the single-target fan following structure of the present invention; Figure 5 This is a schematic diagram of the multi-person air supply range structure of the present invention; Figure 6 This is a schematic diagram of the multi-person airspace on-demand air supply structure of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Please see Figures 1-6 This invention provides a technical solution: a fan adaptive following steering method based on AI decision-making.

[0028] The sensors used in step two include, but are not limited to, cameras, depth sensors, microphone arrays, and infrared pyroelectric detectors. The data in step two includes visual images, depth maps, sound source location, and thermal distribution. In step two, the direction vector includes the human body's position, movement trajectory, identity features, and acoustic and thermal cues. A collaborative representation mechanism is used in step two to maintain spatiotemporal correlation and mutual information, improving the robustness of perception in complex scenarios such as occlusion and low light. In step three, the AI ​​decision-making unit has built-in cross-task understanding and planning capabilities. These capabilities combine scenario-based intent reasoning to determine crowd behavior types and user preset preferences. Through an attention mechanism, weights are dynamically allocated to generate the optimal sequence of action commands. Crowd behavior types include whether people are briefly passing by or stopping to cool off. User preset preferences include single-person locking, timed settings, and energy saving. The sequence of action commands includes the fan's horizontal rotation angle, vertical rotation angle, airflow speed, and duration of stay in the area. The fan modes in step four include single-target continuous tracking mode, multi-user on-demand coverage mode, and intelligent servo standby mode. The single-target continuous tracking mode works as follows: A specific target is set to be tracked, the target's feature ID in the joint vector space is locked, and a closed-loop "sensing-decision-control" tracking process is performed to achieve the following of the moving target. The multi-person on-demand coverage mode works by identifying and marking human body distribution hotspots, and then allowing fans to deliver air only to areas where valid human features are detected in the joint characterization space, quickly skipping unoccupied areas to maximize energy efficiency in areas with people. The intelligent servo standby mode works as follows: when no human feature vector is matched in the joint representation space after a preset time, the system determines that the environment is unoccupied, and the fan autonomously enters a low-power servo state; the sensor continues to collect data, and when human activity features reappear in the joint representation space, it immediately triggers wake-up and resumes the air supply state according to the original settings. The method uses a local chip to run the algorithm. The chip is connected to the target fan and the sensor. The chip has a built-in target detection algorithm that detects all people in a single frame image and marks their positions with bounding boxes. The chip monitors the target within its range in real time, based on pixel coordinates and camera intrinsic parameters. The detected center of the human body in the image is converted into a direction vector emanating from the optical center of the camera, thus obtaining the horizontal rotation angle. Given the camera's horizontal field of view (HFOV), vertical field of view (VFOV), image width (W), and height (H), the formula for calculating the horizontal rotation angle is:

[0029] Given the camera's horizontal field of view (HFOV), vertical field of view (VFOV), image width (W), and height (H), the formula for calculating the vertical rotation angle is:

[0030] The method also includes directional air supply control, the specific process of which is as follows: Visual computing: Detecting the center of the human body and calculating the generated orientation angle. ; Angle mapping: mapping the optical axis direction angle of the camera Converted to the absolute target angle of the fan mechanical shaft The rotation matrix is ​​calibrated based on the camera's installation offset. Control quantity generation: For servos, calculate pulse width or pulse number and direction signal; for DC motors, input parameters into the controller to output PWM duty cycle. Control quantity generation: The drive motor rotates, and a closed-loop system is adopted. The actual angle is fed back by the sensor to form a control closed loop. The user turns on the fan and sets preferences, such as single-user lock, timer, or energy-saving mode. The fan's built-in chip reads the current device parameters, including the horizontal rotation angle. Vertical steering angle The current wind speed setting, sweep mode, and user preset preferences are combined into a fixed-dimensional device state feature vector. This feature vector serves as the initial benchmark for subsequent decisions, and all subsequent features are mapped to the joint vector space.

[0031] After the system starts, the sensors collect data in real time at a fixed frequency, including modal data acquisition: the camera collects image data; the depth sensor collects the distance to the human body in the central area; the microphone array collects sound source data; and the infrared pyroelectric detector collects thermal distribution data. Feature values ​​are extracted from the collected data and mapped to a unified high-dimensional vector space to generate directional information vectors. The directional information includes human position, movement trajectory, identity characteristics, and acoustic and thermal cues. Through a collaborative representation mechanism, the system maintains spatiotemporal correlation and mutual information, improving the robustness of perception in complex scenarios such as occlusion and low light. For example, in low light or partial occlusion, the weight of visual features decreases, while the weight of sound source orientation features automatically increases. This ensures that even when the user is walking, talking, or entering a partially occluded area, the system can still perceive the person through sound or thermal signals and will not lose track of them.

[0032] The direction information vector is concatenated with the current device state vector and input into the AI ​​decision-making unit. The unit analyzes the number of people, dwell time, and movement trajectory to determine the behavior type: short-term passing by or stopping to cool off. Combined with user preset preferences, such as energy saving priority, the unit determines the current task objective. The AI ​​decision-making unit dynamically weights different features. When the light is normal, the visual weight is greater than the acoustic and thermal weights. When the light is dim or obstructed, the visual weight decreases, while the acoustic and thermal weights increase. A sequence of action instructions is generated, including the fan's horizontal turning angle, vertical turning angle, sweeping speed, and dwell time in the area. If a user is walking alone, the fan will follow the movement. If multiple people are distributed in two locations, the fan will turn to the two areas in turn and stay there.

[0033] The AI ​​decision-making unit automatically and seamlessly switches between three modes based on the continuous changes in the direction information vector, and drives the motor to perform steering: like Figure 2 , Figure 3 and Figure 4 As shown, there are two modes: Mode 1 and Single Target Continuous Tracking Mode. Users can set a single-person lock or a scene where only one person stays for more than a set time. The target's feature ID is locked, its position is calculated every frame, and continuous turning commands are generated. Even if the target is briefly obscured, it can still maintain tracking through sound or heat signals. The fan automatically rotates and follows the user as they walk until the target leaves the monitoring range. like Figure 2, Figure 5 and Figure 6 As shown, the second mode and the multi-person on-demand coverage mode are used. When ≥2 people are detected and distributed in different areas, and the user has not set a single person lock, the human distribution hot zone is identified and marked. The fan turns to each hot zone in turn to deliver air and quickly skips the unoccupied area. Mode 3, Intelligent Servo Standby Mode: If no human feature vector is matched in the joint representation space within a preset time, the system determines that the environment is unoccupied, the fan motor stops, and the sensors continue to collect data. When human activity features reappear in the joint representation space, the system immediately triggers wake-up and resumes the air supply state according to the original settings.

[0034] Working principle: The system acquires the current device settings information of the fan, encodes the device settings information into a device state feature vector, which serves as the initial benchmark representation for subsequent AI decision-making and is uniformly incorporated into the joint vector space; it collects data within a set range in real time through sensors, maps each feature value to a unified high-dimensional vector space, and generates a direction information vector; it inputs the direction information and the current fan state feature value into the AI ​​decision-making unit to generate a sequence of action commands; based on the continuous analysis results of the AI ​​decision-making unit, it realizes intelligent steering of the fan.

[0035] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A fan adaptive steering method based on AI decision-making, characterized in that: Includes the following steps: Step 1: Obtain the current device settings information of the fan, encode the device settings information into a device status feature vector, use it as the initial benchmark representation for subsequent AI decision-making, and uniformly incorporate it into the joint vector space; Step 2: Collect data within a set range in real time using sensors, map each feature value to a unified high-dimensional vector space, and generate a direction information vector; Step 3: Input the direction information and the current fan status feature value into the AI ​​decision unit to generate a sequence of action commands; Step 4: Based on the continuous analysis results of the AI ​​decision-making unit, realize the intelligent steering of the fan.

2. The fan adaptive following steering method based on AI decision-making according to claim 1, characterized in that: The sensors in step two include, but are not limited to, cameras, depth sensors, microphone arrays, and infrared pyroelectric detectors. The data in step two includes visual images, depth maps, sound source location, and thermal distribution.

3. The fan adaptive following steering method based on AI decision-making according to claim 1, characterized in that: In step two, the direction vector includes human position, movement trajectory, identity features, and acoustic and thermal cues. In step two, a collaborative representation mechanism is used to maintain spatiotemporal correlation and mutual information, thereby improving the robustness of perception in complex scenarios such as occlusion and low light.

4. The fan adaptive following steering method based on AI decision-making according to claim 1, characterized in that: In step three, the AI ​​decision-making unit has built-in cross-task understanding and planning capabilities. These capabilities combine scenarios to perform intent reasoning, determine the type of crowd behavior, and combine user preset preferences. Through an attention mechanism, weights are dynamically allocated to generate the optimal sequence of action instructions. The type of crowd behavior includes whether the crowd is passing by briefly or staying to cool off. The user preset preferences include single-person locking, timed settings, and energy saving. The sequence of action instructions includes the fan's horizontal turning angle, the fan's vertical turning angle, the sweeping speed, and the duration of stay in the area.

5. The fan adaptive following steering method based on AI decision-making according to claim 1, characterized in that: The fan modes in step four include single-target continuous tracking mode, multi-person on-demand coverage mode, and intelligent servo standby mode.

6. The fan adaptive following steering method based on AI decision-making according to claim 5, characterized in that: The specific content of the single-target continuous tracking mode is as follows: set a specific target to be tracked, lock the feature ID of the target in the joint vector space, and perform "sensing-decision-control" closed-loop tracking to achieve the following of the moving target.

7. The fan adaptive following steering method based on AI decision-making according to claim 5, characterized in that: The specific content of the multi-person on-demand coverage mode is as follows: identify and mark the hot zones of human body distribution, and the fan only delivers air to areas where effective human characteristics are detected in the joint characterization space, quickly skipping unoccupied areas to maximize the energy efficiency of blowing air in areas with people.

8. The fan adaptive following steering method based on AI decision-making according to claim 5, characterized in that: The intelligent servo standby mode is as follows: when no human feature vector is matched in the joint representation space after a preset time, the system determines that there is no one in the environment and the fan enters a low-power servo state autonomously; the sensor continues to collect data, and when human activity features reappear in the joint representation space, it immediately triggers wake-up and restores the air supply state according to the original settings.

9. The fan adaptive following steering method based on AI decision-making according to claim 1, characterized in that: The method uses a local chip to implement the algorithm. The chip is connected to the target fan and the sensor. The chip has a built-in target detection algorithm that detects all people in a single frame image and marks their positions with bounding boxes. The chip monitors the target situation within its real-time range based on pixel coordinates and camera intrinsic parameters. The detected center of the human body in the image is converted into a direction vector emanating from the optical center of the camera, thus obtaining the horizontal rotation angle. Given the camera's horizontal field of view (HFOV), vertical field of view (VFOV), image width (W), and height (H), the formula for calculating the horizontal rotation angle is: Given the camera's horizontal field of view (HFOV), vertical field of view (VFOV), image width (W), and height (H), the formula for calculating the vertical rotation angle is:

10. The fan adaptive following steering method based on AI decision-making according to claim 1, characterized in that: The method also includes directional air supply control, the specific process of which is as follows: Visual computing: Detecting the center of the human body and calculating the generated orientation angle. ; Angle mapping: mapping the optical axis direction angle of the camera Converted to the absolute target angle of the fan mechanical shaft The rotation matrix is ​​calibrated based on the camera's installation offset. Control quantity generation: For servos, calculate pulse width or pulse number and direction signal; for DC motors, input parameters into the controller to output PWM duty cycle. Control quantity generation: The drive motor rotates using a closed-loop system, and the actual angle is fed back by sensors to form a control closed loop.

Images