Intelligent lighting real-time scene perception and decision-making system based on edge calculation

Abstract

The invention discloses an intelligent lighting real-time scene perception and decision-making system based on edge calculation, and belongs to the technical field of intelligent lighting. The system comprises a plurality of intelligent lighting nodes and an edge computing gateway. The edge computing gateway is used for constructing a dynamic vector map to perceive the motion state of a dynamic target in a scene; introducing a trajectory confidence evaluation unit to perform reliability evaluation on the dynamic vector map so as to quantify the predictability of the scene; and conditional prediction and control are carried out based on an evaluation result. According to the method, the technical problems of poor predictive control robustness and easiness in false triggering caused by incapability of effectively distinguishing a reliable target and an interference signal in coping with complex dynamic scenes such as an underground garage in the prior art can be solved. According to the invention, the anti-interference capability and the control reliability of the system in a real and complex environment are improved while the advanced lighting effect of waiting for vehicles by light is realized.

Description

Technical Field

[0001] This invention belongs to the field of intelligent lighting technology, and specifically relates to a real-time scene perception and decision-making system for intelligent lighting based on edge computing. Background Technology

[0002] Edge computing-based smart lighting systems aim to achieve low-latency, high-efficiency lighting control through localized data processing and decision-making, and to save energy in scenarios such as underground parking garages and tunnels through "on-demand lighting" solutions. Existing technologies typically use sensors deployed at the edge to detect the presence of targets in the scene and illuminate the corresponding area's lights when a target is detected. Some improved solutions attempt to introduce simple Doppler frequency shift analysis to predict the speed of high-speed dynamic targets (such as vehicles), hoping to solve the "lights on when a vehicle passes, spatiotemporal mismatch" problem caused by control delays.

[0003] However, when applied to real, complex dynamic scenarios, the aforementioned existing technologies face a more severe challenge in terms of perception robustness. In real environments such as underground parking garages, there are often multiple vehicles converging, pedestrians and vehicles mixing, non-target objects moving (like duct fans), and multipath reflections from walls, creating a variety of complex interferences. Whether based on simple "presence" perception or simple single-target velocity prediction, existing technologies lack a mechanism to assess the overall signal quality and reliability of the scene, failing to effectively distinguish between single, predictable, valid targets and complex, unpredictable, or invalid interference signals. This directly leads to existing predictive lighting systems being highly susceptible to problems such as not responding to real targets or being falsely triggered by interference signals in practical applications, rendering them lacking necessary control reliability and commercial value. Summary of the Invention

[0004] This invention provides a smart lighting real-time scene perception and decision-making system based on edge computing, which aims to solve the technical problems of poor predictive control robustness and susceptibility to false triggering caused by the lack of an evaluation mechanism for the reliability of scene signals when dealing with complex dynamic scenes in existing technologies.

[0005] In view of the above problems, the present invention provides a smart lighting real-time scene perception and decision-making system based on edge computing, including multiple smart lighting nodes and an edge computing gateway. The system includes the edge computing gateway, which constitutes... The dynamic vector map building unit is used to aggregate Doppler radar echo data from multiple smart lighting nodes, extract the motion vectors of dynamic targets, and build a dynamic vector map. The trajectory confidence assessment unit is used to calculate a trajectory confidence score that characterizes the predictability of the dynamic target's motion based on the dynamic vector map. The condition prediction and control unit is used to perform predictive control on the smart lighting node based on the comparison result of the trajectory confidence score and the preset threshold.

[0006] The technical solution provided in this application introduces a "trajectory confidence assessment" decision verification step. Before executing predictive control, the "predictability" of the perceived dynamic scene is quantitatively assessed. Advance control is only executed when the scene is confirmed to be clear and the target reliable, thus constructing a "verify first, predict later" decision-making paradigm. This technical solution has at least the following overall technical effects: while achieving seamless, advanced following illumination for high-speed dynamic targets, it greatly improves the system's anti-interference capability and control reliability in real-world complex environments such as multi-target intersections, mixed pedestrian and vehicle traffic, and clutter interference, avoiding erroneous predictions and control, and achieving a unity of safety, energy saving, and intelligent experience. Attached Figure Description

[0007] Figure 1 This is a system architecture block diagram of a smart lighting real-time scene perception and decision-making system based on edge computing, provided in an embodiment of the present invention. Detailed Implementation

[0008] The above technical solutions will now be described in detail with reference to the accompanying drawings and specific embodiments to provide a better understanding of them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.

[0009] Please see Figure 1 A real-time scene perception and decision-making system for smart lighting based on edge computing includes multiple smart lighting nodes and an edge computing gateway. The system includes the edge computing gateway, which comprises: The dynamic vector map building unit is used to aggregate Doppler radar echo data from multiple smart lighting nodes, extract the motion vectors of dynamic targets, and build a dynamic vector map. The trajectory confidence assessment unit is used to calculate a trajectory confidence score that characterizes the predictability of the dynamic target's motion based on the dynamic vector map. The condition prediction and control unit is used to perform predictive control on the smart lighting node based on the comparison result of the trajectory confidence score and the preset threshold.

[0010] This invention discloses a real-time scene perception and decision-making system for smart lighting based on edge computing. In one specific embodiment, the system operates within an underground parking garage lighting network consisting of multiple smart lighting nodes deployed along the driveway and an edge computing gateway. All core perception, verification, and decision-making computations are implemented by a processor deployed within the edge computing gateway, executing computer programs stored in its memory. When executed, this computer program forms a logically unified decision-making system. The system's entire workflow begins with real-time perception of the dynamic scene. The goal of this perception phase is to construct a dynamic vector map that provides a reliable data foundation for subsequent decision-making.

[0011] To achieve the goal of constructing a dynamic vector map, the decision-making system deployed within the edge computing gateway first aggregates Doppler radar echo data. In this step, the edge computing gateway, acting as the network's data aggregation center, continuously and periodically receives raw echo data streams uploaded from all smart lighting nodes within its coverage area. Each smart lighting node is equipped with a Doppler radar sensor, which continuously emits electromagnetic waves and collects echo signals containing dynamic target information, uploading unprocessed or only pre-filtered raw data frames in real time via its local communication module. The edge computing gateway timestamps and identifies the source node in the received data frames to ensure data synchronization and spatial traceability. As an alternative implementation, smart lighting nodes can also perform partial signal processing locally, uploading only the extracted key feature data to reduce network bandwidth consumption.

[0012] After aggregating the raw echo data from multiple nodes, the decision system then performs motion vector extraction and transformation. The purpose of this step is to convert the echo signals from the physical world into machine-understandable, structured motion vector data. Specifically, the decision system performs a Fast Fourier Transform (FFT) algorithm on each data frame or multiple frames within a time window, transforming it from the time domain to the frequency domain to obtain the Doppler spectrum. The system then searches this spectrum for significant peaks with energy exceeding a preset noise floor; each peak represents a detected dynamic target. For each identified frequency shift peak, the system records its frequency offset. Subsequently, the system followed the physical formulas. Calculate the radial velocity of the target relative to the sensor. Among them, the light-speed C-radar transmission frequency The system has known fixed parameters, and the angle θ between the radar beam and the lane is a preset parameter calibrated during the deployment of the node. Finally, the decision system combines the deployment orientation information of the source node to calculate the radial velocity. It is decomposed into a velocity component and a direction component along the lane direction. These two components together constitute a motion vector containing velocity and direction information.

[0013] After successfully converting the raw echo data into discrete motion vector information in batches, the perception stage enters its final step: the generation of a dynamic vector map. This involves organizing the disordered, discrete vector information into a structured data entity with a spatiotemporal context, capable of being processed by subsequent advanced algorithms. In this embodiment, the dynamic vector map is constructed as a hash table or dictionary structure indexed by node IDs. The decision system traverses all motion vectors generated in the previous step, storing or updating the vector information in real time under the corresponding node ID entry in the hash table based on the source node identifier of each vector. Each entry not only stores the latest motion vector but can also selectively store a historical vector queue containing several past time steps. Thus, the system constructs and continuously maintains a machine-readable dynamic vector map that fully represents the motion state of all dynamic targets in the current scene. The successful generation of this dynamic vector map serves as the input for the next decision verification stage.

[0014] After the dynamic vector map is constructed, the decision-making system enters the trajectory confidence score evaluation and implementation phase: the quality of the motion information contained in the dynamic vector map is assessed to quantify whether the current scene is suitable for reliable trajectory prediction. To achieve this goal, the decision-making system first performs the calculation of intermediate variables, a step aimed at extracting two key indicators from the dynamic vector map that characterize its overall quality.

[0015] To calculate the first intermediate variable, vector field spatial consistency, the decision system traverses all motion vectors recorded in the current dynamic vector map. For each motion vector, the system extracts its velocity and direction values. Subsequently, the system calculates the statistical standard deviation or variance of the velocity values ​​for the entire vector set, and the angular standard deviation of the direction values ​​for all directions. Finally, the system can either merge these two statistics into a single value according to preset weights, or select one as a quantitative indicator characterizing the degree of "orderliness" or "chaos" in the spatial distribution of the vector field; this indicator is vector field spatial consistency. For example, when there is only one steadily moving car in the scene, the motion vectors detected by all nodes in the map are highly similar in velocity and direction, and the calculated velocity and angular standard deviations will approach zero, resulting in a high vector field spatial consistency value. Conversely, if there are multiple intersecting vehicles or a large number of false targets caused by multipath reflections in the scene, the motion vectors will exhibit significant differences, and the calculated standard deviations will be larger, resulting in a lower vector field spatial consistency value.

[0016] To calculate the second intermediate variable, the Doppler signal quality index, the decision system needs to trace back to the original Doppler radar echo data that constitutes the dynamic vector map. The system first identifies the echo signals that contribute most to the calculation of the spatial consistency of the vector field and constitute the main motion trend. For these identified key echo signals, the system calculates their signal-to-noise ratio (SNR), which can be calculated by comparing the peak energy in the signal spectrum with the noise floor energy of the spectrum. Simultaneously, the system can analyze the intensity stability of these signals over multiple consecutive time frames, calculating the amplitude of their jitter or flicker. Finally, the system integrates multiple sub-indicators, such as SNR and stability, using a pre-defined weighted algorithm into a single quantitative index that represents the "reliability" of the signal itself constituting the main motion trend—this index is the Doppler signal quality index.

[0017] After calculating the two intermediate variables, vector field spatial consistency and Doppler signal quality index, the decision system then executes a confidence score generation step based on a fusion formula. The purpose of this step is to integrate two intermediate variables, which have different dimensions but both reflect scene reliability, into a unified and final decision criterion. In this embodiment, the decision system substitutes the two intermediate variables into a fusion formula model consisting of the product of a saturation adjustment function and a threshold gating function. This multiplicative structure ensures that if the evaluation result of either variable is too low, it will significantly lower the final confidence score.

[0018] To achieve the functional structure and technical effects in the fusion formula, the saturation adjustment function is specifically implemented using a hyperbolic tangent function structure. The decision system will ensure consistency across the vector field space. After multiplying by a weighting factor α, it is used as the input to the hyperbolic tangent function, i.e. The hyperbolic tangent function structure ensures that when the vector field space consistency value is already high, its calculation result will smoothly approach the upper limit of 1, rather than growing indefinitely. This avoids the system making overconfident assessments of a scene that is already very clear. The specific implementation of the threshold gating function adopts... The decision-making system will use the Doppler signal quality index. Multiply by a weighting factor β, then take its negative value, and then use that negative value. The exponent of the natural exponential function is used in the calculation. This ensures that when the Doppler signal quality index is low and close to zero, the calculated result is always close to zero. This allows a signal with extremely poor quality to act as a veto gate in subsequent multiplication operations, effectively filtering out interference caused by low-quality clutter signals at the decision-making level.

[0019] After completing the design of the function structure, the decision system finally executes the trajectory confidence score output, i.e. This step provides a specific numerical example to illustrate the entire calculation process: assuming that in a scene, the calculated vector field space is consistent... The Doppler signal quality index is 0.95. The system's preset weighting factors are α = 2.5 and β = 1.5, with a base value of 1.2. The decision system first calculates the saturation adjustment function: Next, we calculate the threshold gating function: Finally, multiply the two results to obtain the final trajectory confidence score. The trajectory confidence score has now been successfully generated, and its value will serve as the input for the next stage of conditional prediction and control.

[0020] After the trajectory confidence score evaluation is completed and a quantified trajectory confidence score is generated, the decision-making system enters its final working phase: the execution and implementation of conditional prediction and control. Based on the evaluation results produced in the decision verification phase, a final, highly robust lighting control decision is made. To achieve this goal, the decision-making system first performs a confidence score gating judgment. This step is a logical decision checkpoint, and its specific implementation process is as follows: the decision-making system compares the real-time calculated trajectory confidence score with a preset execution threshold that can be configured by the user or administrator. The value of this execution threshold, for example, can be set to 0.7, which represents the minimum reliability threshold required by the system for predictive behavior.

[0021] After completing the gating judgment, the decision-making system's behavior will follow two distinct execution paths based on the comparison results. The first path is prediction and proactive control under high confidence. When the trajectory confidence score is higher than the execution threshold, the decision-making system determines that a clear, reliable, and predictable dynamic target has been detected. Under this condition, the system will execute a complete action chain: First, the system extracts the main motion vector, identified during the calculation of vector field spatial consistency and composed of high-quality signals, from the dynamic vector map. Next, based on the direction information of this main motion vector, the system predicts the trajectory of the dynamic target, i.e., the sequence of downstream smart lighting nodes it will pass through. Simultaneously, based on the speed information of this main motion vector and the fixed distance data between each smart lighting node pre-stored in the system, the system calculates the estimated time for the dynamic target to reach each downstream node. Finally, based on the predicted trajectory and time, the decision-making system generates and sends a pre-activation command to one or more downstream smart lighting nodes on the predicted trajectory. The pre-activation command can include specific control parameters, such as target brightness, color temperature, and a dynamic fading time associated with the predicted arrival time, to achieve a smooth lighting transition.

[0022] The second execution path is suppression and safe mode rollback under low confidence. When the trajectory confidence score is not higher than the execution threshold, the decision system determines that the current scene is too complex, there is too much signal interference, or there is no clear prediction target, making it unsuitable for proactive control. Under this condition, the decision system will actively suppress any predictive control behavior, i.e., it will neither generate nor send any pre-activation commands. Simultaneously, to ensure basic lighting functions and safety, the decision system will instruct the entire lighting network to roll back to a preset safe operating mode. In one embodiment, this safe operating mode can be a basic, sensor-based, instantaneous lighting mode, where the lights at a node are only triggered when a vehicle actually arrives below it. In another embodiment, this safe operating mode can also be maintaining a constant, low-brightness, safe basic lighting covering all areas. This suppression and rollback mechanism ensures that the system will not produce chaotic lighting control due to erroneous predictions when facing uncertainty, guaranteeing the overall robustness of the system.

[0023] The embodiments described above are only some embodiments of the present invention, and not all embodiments. It should be noted that the present invention is not limited to the embodiments described above. Any modifications, equivalent substitutions, variations, or improvements made by those skilled in the art based on the above embodiments without creative effort, within the scope of the concept and principles of the present invention, should fall within the protection scope claimed by the claims of the present invention.

Claims

1. A real-time scene perception and decision-making system for smart lighting based on edge computing, comprising multiple smart lighting nodes and an edge computing gateway, characterized in that, The edge computing gateway includes: The dynamic vector map building unit is used to aggregate Doppler radar echo data from multiple smart lighting nodes, extract the motion vectors of dynamic targets, and build a dynamic vector map. The trajectory confidence assessment unit is used to calculate a trajectory confidence score that characterizes the predictability of the dynamic target's motion based on the dynamic vector map. The condition prediction and control unit is used to perform predictive control on the smart lighting node based on the comparison result of the trajectory confidence score and the preset threshold.

2. The smart lighting real-time scene perception and decision-making system based on edge computing according to claim 1, characterized in that, The dynamic vector map construction unit is specifically used to: perform fast Fourier transform processing on the Doppler radar echo data to extract the Doppler frequency shift component, and convert the Doppler frequency shift component into the motion vector of the dynamic target.

3. The smart lighting real-time scene perception and decision-making system based on edge computing according to claim 1, characterized in that, The trajectory confidence evaluation unit is specifically used for: Based on the dynamic vector map, the spatial consistency of the vector field is determined; Based on the Doppler radar echo data that constitutes the dynamic vector map, the Doppler signal quality index is determined. The trajectory confidence score is calculated based on the spatial consistency of the vector field and the Doppler signal quality index.

4. The smart lighting real-time scene perception and decision-making system based on edge computing according to claim 3, characterized in that, The trajectory confidence evaluation unit calculates the trajectory confidence score by multiplying a saturation adjustment function, which characterizes the spatial consistency of the vector field, with a threshold gating function, which characterizes the Doppler signal quality index.

5. The smart lighting real-time scene perception and decision-making system based on edge computing according to claim 4, characterized in that, The saturation adjustment function adopts a hyperbolic tangent function structure so that its contribution to the trajectory confidence score tends to the upper limit as the consistency of the vector field space increases.

6. The smart lighting real-time scene perception and decision-making system based on edge computing according to claim 4, characterized in that, The threshold gating function adopts The mathematical structure in which The Doppler signal quality index is β, which is a preset weighting factor, so that when the Doppler signal quality index is below a critical value, its contribution to the trajectory confidence score is close to zero.

7. The smart lighting real-time scene perception and decision-making system based on edge computing according to claim 6, characterized in that, The trajectory confidence assessment unit calculates the trajectory confidence score (STC) using the following formula: in, The vector field space consistency is defined by α, which is a preset weighting factor.

8. The smart lighting real-time scene perception and decision-making system based on edge computing according to claim 1, characterized in that, The condition prediction and control unit is specifically used for: Determine whether the trajectory confidence score is higher than a preset execution threshold; If it is higher, it is considered that a dynamic target that can be predicted and controlled has been perceived; If the value is not higher, then predictive control of the smart lighting node is suppressed.

9. A smart lighting real-time scene perception and decision-making system based on edge computing according to claim 8, characterized in that, After identifying that the dynamic target can be predicted and controlled, the condition prediction and control unit is specifically used to: extract the main motion vector from the dynamic vector map, predict the motion trajectory and arrival time of the dynamic target based on the main motion vector, and send a pre-turn-on command to the smart lighting node on the motion trajectory.

10. A smart lighting real-time scene perception and decision-making system based on edge computing according to claim 8, characterized in that, When suppressing the predictive control, the condition prediction and control unit instructs the system to revert to an instant lighting mode based on sensor presence detection or a basic lighting mode.

Images