Motor vehicle light irradiation range detection method based on laser radar

Abstract

The invention discloses a motor vehicle light irradiation range detection method based on a laser radar, and relates to the technical field of motor vehicle intelligent illumination, and the method comprises the steps: obtaining initial environment perception reference information through a perception unit of a vehicle; inputting the initial environment perception reference information into a reinforcement learning decision engine, and outputting a high-value detection point set based on an information gain maximization strategy; controlling the laser radar to execute directional scanning according to the high-value detection point set, controlling the to-be-detected vehicle lamp unit to perform concurrent modulation luminescence according to a pre-distributed quasi-orthogonal pseudo-random spreading code sequence, and collecting a mixed echo signal sequence of coding characteristics of each vehicle lamp unit; and separating an independent optical signal component corresponding to the quasi-orthogonal pseudo-random spreading code sequence from the mixed echo signal sequence based on a code pattern coherent resonance principle. According to the invention, accurate decoupling and real-time evaluation of space energy distribution of each independent light source in a multi-light-source complex scene are realized, and light source level fine control capability is provided for a vehicle light unit.

Description

Technical Field

[0001] This invention relates to the field of intelligent lighting technology for motor vehicles, and in particular to a method for detecting the illumination range of motor vehicle headlights based on lidar. Background Technology

[0002] In the field of intelligent lighting technology for motor vehicles, existing technologies mainly rely on on-board cameras to collect images of the light field ahead and analyze the lighting effect through image processing algorithms. For example, deep learning models are used to identify the cut-off line between light and dark and compare it with a standard light pattern template. This type of solution has been widely used in adaptive high beam systems, which can reuse existing vision hardware and has good cost-effectiveness.

[0003] Existing technologies have limitations when dealing with complex lighting scenarios with multiple light sources: the two-dimensional images captured by cameras are essentially superimposed projections of the contributions of each light source, making it difficult to accurately decouple the spatial energy distribution of each independent light source in dynamic scenes. Especially under conditions of vehicle posture changes such as curves and slopes, the illumination angle and relative position of each light source unit are constantly changing. Traditional methods cannot distinguish the specific contributions of different light sources to specific areas, resulting in biases in the evaluation of the overall lighting effect and failing to provide sufficiently accurate data support for fine control at the light source level. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for detecting the illumination range of motor vehicle lights based on lidar, which solves the technical problem of difficulty in analyzing the spatial energy distribution of each independent light source in dynamic multi-light source scenes.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for detecting the illumination range of motor vehicle lights based on lidar, which includes acquiring initial environmental perception reference information through the vehicle's sensing unit; The initial environmental perception baseline information is input into the reinforcement learning decision engine, and a set of high-value detection points is output based on the information gain maximization strategy. The lidar is controlled to perform directional scanning based on the high-value detection point set, and the vehicle lamp unit under test is controlled to perform concurrent modulation and emission according to the pre-allocated quasi-orthogonal pseudo-random spreading code sequence, and the mixed echo signal sequence of the coding characteristics of each vehicle lamp unit is collected. Based on the principle of code coherent resonance, independent optical signal components corresponding to quasi-orthogonal pseudo-random spreading code sequences are separated from the mixed echo signal sequence to generate a separate optical contribution point cloud; A comprehensive three-dimensional light field distribution is generated by spatial interpolation and superposition reconstruction based on the separated light contribution point cloud. The three-dimensional light field distribution is compared with the pre-stored regulatory light pattern parameters and dynamic target spatial information to output vehicle lighting adaptive control commands.

[0007] As a preferred embodiment of the vehicle headlight illumination range detection method based on lidar described in this invention, the method includes the following steps: obtaining an initial environmental perception reference through the vehicle's sensing unit. The vehicle's front area is scanned by LiDAR to generate initial point cloud data with three-dimensional coordinates and reflection intensity. Visible light image data of the front environment is collected by the vehicle's onboard camera. The relative speed and azimuth data of moving targets ahead are detected by millimeter-wave radar, and the pitch and yaw angle data of the vehicle are measured in real time using an inertial measurement unit. The initial point cloud data, visible light image data, relative velocity and azimuth data, and pitch and yaw angle data are input into the multi-sensor fusion algorithm to output the initial environmental perception reference information.

[0008] As a preferred embodiment of the vehicle headlight illumination range detection method based on lidar described in this invention, the method includes the following steps: inputting an initial environmental perception benchmark into a reinforcement learning decision engine, and outputting a high-value detection point set based on an information gain maximization strategy. The initial environmental perception benchmark is used to extract multi-dimensional spatial light field uncertainty features through a spatiotemporal attention encoder to form a state representation vector; The state representation vector is input into the light field exploration network of the reinforcement learning decision engine. Based on the information gain maximization strategy, Monte Carlo tree search is performed in the potential probe point space to obtain the expected information gain value. Potential probe points whose expected information gain values ​​conform to the dynamic Pareto front are selected to form a candidate probe point set. Spatial coverage optimization based on the Voronoi diagram is performed on the candidate probe point set to form a high-value probe point set.

[0009] As a preferred embodiment of the vehicle headlight illumination range detection method based on lidar described in this invention, the method includes the following steps: controlling the lidar to perform directional scanning based on a high-value detection point set, and controlling the headlight unit under test to perform concurrent modulation emission according to a pre-allocated quasi-orthogonal pseudo-random spreading code sequence: The high-value detection point set is converted into lidar scanning control commands, which drive the lidar to perform directional scanning. The quasi-orthogonal pseudo-random spreading code sequence is converted into a vehicle lamp unit modulation control command. The vehicle lamp unit modulation control command drives the vehicle lamp unit under test to perform concurrent modulation and emission. The directional scanning and concurrent modulation and emission are executed in time-aligned. The lidar receives reflected signals from the detection area, forming a mixed echo signal sequence.

[0010] As a preferred embodiment of the vehicle headlight illumination range detection method based on lidar described in this invention, the method includes the following steps: Collecting the mixed echo signal sequence of the coded features of each headlight unit. The concurrently modulated light signal is reflected by the object surface within the detection area, and the reflected light signal and the ambient background light form an optical superposition during spatial propagation. The optical superposition signal is received by the photodetector of the lidar and converted into an electrical signal. The electrical signal is amplified by a transimpedance amplifier to form an analog signal. The analog signal is digitized by a high-speed analog-to-digital converter at a sampling rate. The digitized signal is then segmented and organized according to the detection points to form a mixed echo signal sequence.

[0011] As a preferred embodiment of the vehicle headlight illumination range detection method based on lidar described in this invention, the method includes the following steps: separating independent optical signal components corresponding to the quasi-orthogonal pseudo-random spreading code sequence from the mixed echo signal sequence based on the code pattern coherent resonance principle to generate a separated optical contribution point cloud: The hybrid echo signal sequence and the quasi-orthogonal pseudo-random spreading code sequence are matched in a chaotic synchronous resonance in the multi-scale time-frequency domain to generate a lock signal. The locking signal is separated into the chaotic attractor signal components corresponding to each code sequence through a self-organizing resonant network. The chaotic attractor signal components are stabilized by Lyapunov exponential stabilization to restore the independent optical signal components of each vehicle lamp unit. The independent optical signal components are tensor-fused with the spatial coordinates of the corresponding detection points to form a separate optical contribution point cloud.

[0012] As a preferred embodiment of the vehicle headlight illumination range detection method based on lidar described in this invention, the method involves: generating a comprehensive three-dimensional light field distribution based on the spatial interpolation and superposition reconstruction of a separate light contribution point cloud, including the following steps: The separated light contribution point cloud is used to generate an independent three-dimensional light field distribution for each headlight unit through the Kriging space interpolation algorithm. The independent three-dimensional light field distributions of each vehicle headlight unit are algebraically superimposed on a three-dimensional spatial grid to generate an algebraic superposition result; Based on the algebraic superposition result, a preliminary three-dimensional light field distribution is generated after Gaussian smoothing filtering. The preliminary three-dimensional light field distribution is corrected by coordinate transformation with the vehicle attitude parameters. The result after coordinate transformation correction is normalized by light intensity to generate a comprehensive three-dimensional light field distribution.

[0013] As a preferred embodiment of the vehicle headlight illumination range detection method based on lidar described in this invention, the method involves comparing the three-dimensional light field distribution with pre-stored regulatory light pattern parameters and dynamic target spatial information to output vehicle headlight adaptive control commands, including the following steps: The three-dimensional light field distribution is projected onto a standard test screen to generate actual light pattern parameters; The difference between the actual beam pattern parameters and the pre-stored regulatory beam pattern parameters is calculated to generate a compliance deviation matrix; Spatial intersection detection is performed between the three-dimensional light field distribution and the dynamic target spatial information to generate the target illuminated area. The glare risk assessment results are generated by comparing the light intensity of the target illuminated area with the glare threshold. The compliance deviation matrix and glare risk assessment results are input into the fuzzy logic controller, which outputs vehicle lighting adaptive control commands.

[0014] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the method for detecting the illumination range of motor vehicle lights based on lidar as described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for detecting the illumination range of motor vehicle lights based on lidar as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By combining directional scanning of lidar with concurrent coding and modulation of the vehicle lighting unit, and using quasi-orthogonal pseudo-random spreading code sequences to uniquely identify each vehicle lighting unit, the independent light contribution of each light source is separated from the mixed echo signal, and a separate light contribution point cloud is constructed. Then, a comprehensive three-dimensional light field distribution is generated through spatial interpolation and superposition reconstruction. This distribution is compared with the regulatory light pattern parameters and dynamic target information, and adaptive control commands are output. This achieves accurate decoupling and real-time evaluation of the spatial energy distribution of each independent light source in complex multi-light source scenarios, providing the vehicle lighting unit with light source-level fine control capabilities. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1This is a flowchart of a method for detecting the illumination range of motor vehicle lights based on lidar.

[0019] Figure 2 This is a flowchart illustrating the adaptive control command for vehicle lighting. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The appearance of an embodiment in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0023] Reference Figures 1-2 This is one embodiment of the present invention, which provides a method for detecting the illumination range of motor vehicle headlights based on lidar, including the following steps: S1. Obtain the initial environmental perception baseline through the vehicle's perception unit.

[0024] S1.1. The area in front of the vehicle is scanned by LiDAR to generate initial point cloud data with three-dimensional coordinates and reflection intensity, and visible light image data of the environment in front is collected by the vehicle-mounted camera.

[0025] Furthermore, by scanning the area in front of the vehicle with LiDAR, initial point cloud data containing three-dimensional coordinates and reflection intensity is generated. This initial point cloud data provides the precise geometric structure and surface reflection characteristics of the scene. At the same time, visible light image data of the environment in front is collected using an onboard camera. The visible light image data provides rich texture and color information.

[0026] Specifically, by combining the precise three-dimensional geometric perception capability of LiDAR with the rich semantic information acquisition capability of cameras, the initial point cloud data of LiDAR can correct the geometric errors caused by lens distortion or perspective projection of cameras, while the visible light image data of vehicle cameras can provide semantic labels and contextual information for the initial point cloud data, such as distinguishing lane lines, traffic signs and vegetation. The simultaneous acquisition of multimodal data uses the depth information of LiDAR to establish an accurate spatial reference frame for visual perception, and at the same time uses the semantic information of vision to interpret the object category represented by the reflection intensity of LiDAR, so that the generated initial point cloud data and visible light image data are strictly registered in space and complement each other in information.

[0027] S1.2 Detect the relative speed and azimuth data of moving targets ahead using millimeter-wave radar, and use an inertial measurement unit to measure the vehicle's pitch and yaw angles in real time.

[0028] Furthermore, the relative velocity and azimuth data of moving targets ahead are detected by millimeter-wave radar. This relative velocity and azimuth data provides precise radial velocity and angular information of the moving targets. At the same time, the pitch and yaw angle data of the vehicle are measured in real time using an inertial measurement unit. The pitch and yaw angle data describe the dynamic attitude changes of the vehicle itself, thus improving the environmental state description from both dynamic and static dimensions: the relative velocity and azimuth data provided by millimeter-wave radar are specifically for dynamic targets, and its Doppler effect can provide velocity vectors that are difficult for lidar and cameras to obtain directly; while the pitch and yaw angle data provided by the inertial measurement unit characterize the motion state of the platform, i.e., the vehicle itself, which is crucial for correcting the offset of other sensor data caused by changes in vehicle attitude.

[0029] Specifically, the observation of external dynamic targets is separated from the monitoring of its own motion state. The millimeter-wave radar focuses on the accurate acquisition of the motion parameters of the external target, while the inertial measurement unit focuses on the independent measurement of its own motion attitude. The data of the two are independent of each other and mutually verify each other, thus constructing a descriptive framework that includes the complete dynamic relationship between its own motion and the motion of the external target.

[0030] S1.3 Input the initial point cloud data, visible light image data, relative velocity and azimuth data, and pitch and yaw angle data into the multi-sensor fusion algorithm to output the initial environmental perception reference information.

[0031] Furthermore, the initial point cloud data, visible light image data, relative velocity and azimuth data, and pitch and yaw angle data are input into the multi-sensor fusion algorithm. Based on timestamps and spatial coordinate systems, all input data are spatiotemporally synchronized and coordinate unified. For example, the relative velocity and azimuth data detected by millimeter-wave radar and the visible light image data from the vehicle camera are converted to a unified reference system based on the coordinate system of the initial point cloud data of the lidar. The pitch and yaw angle data of the inertial measurement unit are used to perform motion compensation for the distortion of the scanned point cloud caused by vehicle movement and the motion between image frames. The multi-sensor fusion algorithm performs data association.

[0032] Specifically, for example, the moving target points detected by millimeter-wave radar are associated with clusters in the initial point cloud data of lidar. The associated targets are then matched with detection boxes in the visible light image data of the vehicle camera, assigning visual appearance attributes to the targets. For each tracked target or static element, the multi-sensor fusion algorithm, based on Kalman filtering or similar state estimation algorithms, integrates its precise position and shape from the initial point cloud data, category and texture semantics from the visible light image data, motion speed from relative velocity and azimuth data, and platform motion corrections derived from pitch and yaw angle data to make the optimal estimate of the target's state. The multi-sensor fusion algorithm outputs an initial environmental perception benchmark information that integrates the high-precision three-dimensional structure of the static environment, the precise motion state and classification semantics of the dynamic target, and the real-time attitude of the vehicle itself.

[0033] S2. Input the initial environmental perception benchmark into the reinforcement learning decision engine, and output a set of high-value detection points based on the information gain maximization strategy.

[0034] S2.1 The initial environmental perception benchmark is used to extract the uncertainty features of the multidimensional spatial light field through a spatiotemporal attention encoder to form a state representation vector.

[0035] Furthermore, the initial environmental perception baseline information is fed into the spatiotemporal attention encoder as input. The spatiotemporal attention encoder extracts spatial feature maps from the initial environmental perception baseline information through a convolutional neural network. These spatial feature maps capture static spatial patterns such as illumination intensity, object contours, and material reflection characteristics in different regions. The spatiotemporal attention encoder concatenates and aligns the spatial feature map of the current frame with feature maps from multiple consecutive frames in the historical buffer in the temporal dimension, and uses a recurrent neural network or a temporal convolutional network to model the dynamic evolution of the light field features.

[0036] Specifically, for example, the moving trajectory of the taillights of a moving vehicle or the gradual change in the illumination range of a street lamp, the spatiotemporal attention encoder uses a multi-head self-attention mechanism to process the spatiotemporal feature sequence. The calculation of attention weights is based not only on the intensity of the feature itself, but more importantly on the variance of the feature in the time series, the difference from the features of neighboring regions, and the confidence level calculated by the sensor noise model. This dynamically highlights spatiotemporal regions with drastic feature changes, inconsistencies with the surrounding environment, or low confidence levels. These regions are identified as regions with high light field uncertainty. The spatiotemporal attention encoder performs global pooling and linear projection on all spatiotemporal features after attention weighting, compressing them into a fixed-dimensional state representation vector. This state representation vector integrates the deterministic layout and uncertain distribution of the ambient light field. The spatiotemporal attention encoder encodes the uncertainty of the perception process itself as a core feature.

[0037] S2.2 Input the state representation vector into the light field exploration network of the reinforcement learning decision engine, perform Monte Carlo tree search in the potential probe point space based on the information gain maximization strategy to obtain the expected information gain value, and select potential probe points whose expected information gain values ​​conform to the dynamic Pareto front to form a candidate probe point set.

[0038] Furthermore, the state representation vector is input into the light field exploration network of the reinforcement learning decision engine. The light field exploration network is a deep neural network that receives the state representation vector and outputs a preliminary value assessment in the potential probe space, initiating the Monte Carlo tree search process. Each simulation of the Monte Carlo tree search begins with the current state representation vector. The light field exploration network selects potential probes for each simulation step and predicts the newly generated state representation vector and corresponding information gain after the probe is performed based on a simplified light field propagation and update model. The information gain is calculated based on the change in the uncertainty feature components in the state representation vector, quantified by comparing the entropy reduction of the uncertainty features before and after the probe. After numerous simulations, each potential probe accumulates an average expected information gain value after backpropagation.

[0039] Specifically, all evaluated potential probe points form a point set on a two-dimensional plane based on their expected information gain and estimated probe cost. A non-dominated sorting algorithm is used to identify the Pareto front in this point set. This Pareto front is dynamic, and its shape depends on the environmental complexity implied by the current state representation vector. Potential probe points located on the dynamic Pareto front are selected to form a candidate probe point set. The points in the candidate probe point set represent the probe selection scheme that achieves the optimal trade-off between information gain and probe cost under a specific environmental state.

[0040] S2.3 Perform spatial coverage optimization based on the Voronoi diagram on the candidate probe point set to form a high-value probe point set.

[0041] Furthermore, performing spatial coverage optimization based on the Voronoi diagram on the candidate detection point set begins by using each point in the candidate detection point set as a generator to construct a Voronoi diagram within a two-dimensional or three-dimensional detection space defined by the vehicle position and detection range. The Voronoi diagram divides the space into multiple Voronoi cells, each containing all points in the space whose distance to its corresponding generator is less than its distance to any other generator. Subsequently, the area or volume of each Voronoi cell is obtained. Voronoi cells with excessively large areas or volumes indicate that their corresponding generators are too spatially isolated and not effectively covered by the candidate probe point set. The optimization process is iterative. For Voronoi cells whose areas or volumes exceed the adaptive threshold, the optimization algorithm evaluates the impact of introducing a new probe point in the Voronoi cell on the expected information gain value. If a suitable new point can be found without significantly reducing the expected information gain value, the point is added. For two or more generators whose Voronoi cell areas or volumes are too small due to their proximity, the optimization algorithm considers merging them or removing some points to reduce redundancy. After multiple rounds of iterative adjustments, an optimized point set with a more uniform spatial distribution and controlled size of each Voronoi cell is finally obtained.

[0042] Specifically, the requirements of information theory value and geometric coverage, two different dimensions, are collaboratively optimized within a unified geometric framework. For example, candidate detection point sets may cluster in the central area of ​​the road due to the presence of multiple high-uncertainty vehicle taillights ahead. Although this results in high information gain, it may lead to complete neglect of pedestrian or traffic sign areas on the roadside. Optimization based on Voronoi diagrams can identify these coverage gaps and attempt to introduce detection points into these gaps, even if the direct information gain of the new points is not the highest. Conversely, for points with extremely high information gain and very close proximity, optimization may merge them, as a single detection action may have already acquired most of the information, and the saved resources can be used to cover other areas. This ensures that the high-value detection point set is not only a set of information hotspots but also constitutes a detection network for key spatial areas, ensuring that no important areas are completely ignored.

[0043] S3. Control the lidar to perform directional scanning based on the high-value detection point set, and control the vehicle headlight unit under test to perform concurrent modulation and emission according to the pre-allocated quasi-orthogonal pseudo-random spreading code sequence.

[0044] S3.1. Convert the high-value detection point set into lidar scanning control commands, and drive the lidar to perform directional scanning.

[0045] Furthermore, the process of converting the high-value detection point set into lidar scanning control commands involves mapping spatial coordinates to mechanical or optical scanning actions. Each three-dimensional coordinate point in the high-value detection point set is first resolved into specific azimuth and pitch angle pairs based on the transformation relationship between the lidar's internal coordinate system and the vehicle's global coordinate system. The angle parameters are further arranged into an ordered command sequence. The command sequence not only includes the pointing angle of each target point, but also accurately calculates and inserts the turning time and settling time required to move from one detection point to the next based on the lidar's pulse repetition frequency and the maximum angular velocity of the scanning mechanism. This generates a time-sequential lidar scanning control command sequence that balances scanning accuracy and efficiency. The lidar scanning control commands are sent to the lidar's scanning execution unit. The scanning execution unit sequentially and precisely aligns the laser beam's emission axis with each high-value detection point according to the command sequence and triggers laser pulse emission at the specified settling moment, achieving rapid and accurate directional scanning of the spatial position defined by the high-value detection point set.

[0046] Specifically, the high-value detection point set is an intelligent selection based on the reinforcement learning decision engine's output of maximizing information gain and optimizing spatial coverage, rather than a pre-defined regularized grid. This process converts the high-value detection point set into LiDAR scanning control commands, optimizing it into an optimal scanning trajectory that takes into account physical constraints. For example, if the high-value detection point set contains two points located far to the left and near to the right, the conversion process not only calculates the angles pointing to these two points but also plans the optimal path for the LiDAR galvanometer or rotating mirror to smoothly move from one point to another with minimal time cost and to avoid mechanical overshoot. It also precisely calculates how long it needs to stabilize at each point to ensure a sufficient signal-to-noise ratio, making the scanning action itself part of the intelligent decision-making process. The scanning path is dynamic and adaptive, directly serving the fundamental goal of maximizing information acquisition, rather than executing a fixed scanning pattern that may contain a large number of redundant or missed points.

[0047] S3.2 The quasi-orthogonal pseudo-random spreading code sequence is converted into a vehicle lamp unit modulation control command. The vehicle lamp unit modulation control command drives the vehicle lamp unit under test to perform concurrent modulation and emission. The directional scanning and concurrent modulation and emission are executed in time-aligned.

[0048] Furthermore, the process of converting the quasi-orthogonal pseudo-random spreading code sequence into a lamp unit modulation control command is a digital signal to analog drive signal generation process. The quasi-orthogonal pseudo-random spreading code sequence pre-assigned to each lamp unit under test is first fed into a waveform generator. The waveform generator maps the digital symbol sequence to a specific waveform, for example, mapping symbol 1 to a high-level pulse and symbol 0 to a low-level pulse or a pulse of a specific shape, thereby generating a time-continuous voltage or current reference waveform, i.e., the lamp unit modulation control command. The lamp unit modulation control command is simultaneously sent to the drive circuits of all lamp units under test.

[0049] Specifically, the driving circuit, based on the instantaneous voltage or current reference value of the received vehicle lamp unit modulation control command, amplifies and adjusts the power to precisely control the current flowing through the light-emitting element (such as the LED chip) of the vehicle lamp unit in real time. This ensures that the light output intensity of the vehicle lamp unit strictly follows the waveform changes of the vehicle lamp unit modulation control command, thereby enabling all the vehicle lamp units under test to perform concurrent but different modulated light emission according to their unique quasi-orthogonal pseudo-random spreading code sequences. The time-aligned execution of directional scanning and concurrent modulation emission is achieved through a high-precision synchronous clock source. This clock source provides a unified time base for generating the laser emission timestamp in the lidar scanning control command and the symbol clock for generating the vehicle lamp unit modulation control command, ensuring that the precise moment of each laser pulse emission corresponds to a specific phase on the waveform of the vehicle lamp unit modulation control command. This ensures that the active interrogation of the lidar and the coded response of the vehicle lamp array are strictly synchronized in time.

[0050] S3.3 The lidar receives reflected signals from the detection area and forms a mixed echo signal sequence.

[0051] Furthermore, the process by which a lidar receives reflected signals from the detection area and forms a mixed echo signal sequence is a photoelectric conversion and signal acquisition process. The reflected light signal returning from the detection area is collected by the lidar's optical receiving antenna and focused onto the photosensitive surface of a photodetector, such as an avalanche photodiode. The photodetector converts the incident photon stream into a weak current signal proportional to the optical power. This current signal is first converted into a voltage signal by a transimpedance amplifier and then preliminarily amplified to improve the signal-to-noise ratio and match the input range of subsequent circuits.

[0052] Specifically, the amplified analog voltage signal is then fed into a high-speed analog-to-digital converter (ADC). The ADC samples and quantizes the analog voltage signal uniformly at a pre-set sampling rate, much higher than the modulation control command code rate of the headlight unit and the laser pulse repetition frequency, converting it into a discrete digital signal sequence. This digital signal sequence is a hybrid echo signal sequence, where each sampling point contains not only the strong echo signal generated by the backscattering of the laser pulse at the corresponding moment (when present), but also the background light signal emitted by all concurrently modulated headlight units at that moment, reflected by the environment and entering the receiving antenna, as well as components such as ambient natural light and detector noise.

[0053] S4. Collect the mixed echo signal sequence of the coding characteristics of each vehicle lamp unit.

[0054] S4.1 The concurrently modulated light signal is reflected by the object surface within the detection area, and the reflected light signal and the ambient background light form an optical superposition during spatial propagation.

[0055] Furthermore, the modulated light signal is reflected from the object surface within the detection area. The reflection process follows optical laws such as Lambertian reflection or specular reflection. The intensity of the reflected light signal depends on the reflectivity of the object surface, the angle of incidence, and the instantaneous luminous intensity defined by the modulation control command of the vehicle lamp unit. During its propagation in space, the reflected light signal undergoes incoherent optical superposition with ambient background light such as sunlight, streetlights, or other unmodulated vehicle lights. The optically superimposed signal is a linear sum of multiple light signals in the optical power dimension. Its instantaneous power is equal to the sum of the light power of each independent light source after reflection through different paths to the receiving point. This sum is a continuous function that varies with time, and its variation is determined by the linear combination of the quasi-orthogonal pseudo-random spreading code sequences of each vehicle lamp unit.

[0056] Specifically, ambient light is redefined from interference noise as a structured information carrier. The linear properties of optical superposition are used as the physical basis for lossless information mixing. By actively controlling the vehicle headlight units to perform concurrent modulation and emission, the main ambient light component, vehicle headlights, becomes controllable and known. Its reflected signal is no longer random noise, but a deterministic signal carrying information about the light source's identity and intensity. The linear properties of optical superposition are key. The total optical power reaching the detector is a simple sum of the contributions from each light source. For example, the coded light from the left low beam is reflected from the road surface, and the coded light from the right low beam is reflected from the taillights of the vehicle in front. These two are superimposed on the natural light at the detector. The total light intensity waveform is a weighted sum of these three signals. Since natural light changes relatively slowly, while the coded light from the vehicle headlights changes rapidly and its pattern is known, it is possible to extract the coded light component from the superimposed signal. Instead of attempting to counteract or eliminate ambient light, the system actively modifies the characteristics of some ambient light, transforming it into a beneficial signal. This transforms one of the biggest sources of interference in traditional perception into an information source, improving the robustness and information acquisition capabilities of the perception system in complex lighting environments.

[0057] S4.2 The optical superposition signal is received by the photodetector of the lidar and converted into an electrical signal. The electrical signal is amplified by a transimpedance amplifier to form an analog signal.

[0058] Furthermore, the optical superimposed signal is received by the photodetector of the lidar. The photosensitive surface of the photodetector receives light from all directions within the field of view, converting the instantaneous value of the total incident optical power into the corresponding photocurrent. The amplitude of the photocurrent is proportional to the optical power of the optical superimposed signal, and the changing waveform faithfully reproduces the time-varying characteristics of the optical superimposed signal. The weak photocurrent signal output by the photodetector is then fed into a transimpedance amplifier. The transimpedance amplifier converts the current signal into a voltage signal, providing sufficient gain to increase the signal amplitude so that it can reach the level range that subsequent circuits can process. The design of the transimpedance amplifier needs to consider bandwidth, noise, and dynamic range to ensure that it can amplify the wide-spectrum signal composed of fast laser pulses and relatively slow coded optical modulation without distortion. The voltage signal formed after amplification by the transimpedance amplifier is the analog signal, which is the continuous-time representation of the optical superimposed signal in the electrical domain.

[0059] Specifically, the broadband, high dynamic range optoelectronic receiving front-end based on nanosecond-level laser pulse response and microsecond-level coded optical modulation breaks the limitation of traditional receiving circuits that are usually optimized for a single type of signal. The transimpedance amplifier needs to have a sufficiently low-frequency response to retain the information of the coded optical modulation, while also having a sufficiently high bandwidth to capture the laser pulse. The receiving front-end is regarded as a broadband signal acquisition channel rather than a pulse detector. For example, when receiving a laser pulse echo, the transimpedance amplifier needs to respond quickly to generate a voltage spike. During the pulse interval, the voltage baseline of the amplifier output is not a fixed DC, but fluctuates slowly with the intensity of the coded reflected light. The slow fluctuation carries the key information. The gain and bandwidth design of the transimpedance amplifier need to ensure that it can amplify the weak pulse signal without over-amplifying the DC or low-frequency components caused by strong background light, which would cause saturation. This allows a single receiving channel to simultaneously capture two types of signals with very different time-varying characteristics with high fidelity: transient laser echo and continuous coded optical modulation.

[0060] S4.3 The analog signal is digitized by a high-speed analog-to-digital converter at a sampling rate. The digitized signal is segmented and organized according to the detection points to form a mixed echo signal sequence.

[0061] Furthermore, the analog signal is digitized by a high-speed analog-to-digital converter (ADC) at a sampling rate. The ADC samples the continuous analog voltage signal at uniform time intervals and quantizes the analog voltage value corresponding to each sampling point into a discrete digital code. The sampling rate must satisfy the Nyquist sampling theorem and be more than twice the highest frequency component in the analog signal to ensure distortion-free sampling of the encoded modulation information. The resulting discrete digital sequence is then segmented and organized according to the detection points. The segmentation is based on the precise timestamp of each laser pulse emission recorded in the lidar scanning control command. A time window is taken before and after each timestamp, and all digital sampling points within that window are extracted to form a data segment corresponding to that detection point.

[0062] Specifically, since the echo of each laser pulse appears at a specific distance-related delay after its emission timestamp, this data segment contains both the signal characteristics generated by the laser pulse echo and the signal characteristics generated by the coded background light during a period before and after the laser pulse emission. By arranging the data segments corresponding to all high-value detection points in the scanning order, a hybrid echo signal sequence is finally formed. The hybrid echo signal sequence is a structured dataset in which each element corresponds to a spatial detection point, and each element contains a time sequence, which is a digital record of the electrical signal of the hybrid light field at that point in the time domain.

[0063] S5. Based on the principle of coherent resonance of code patterns, separate the independent optical signal components corresponding to the quasi-orthogonal pseudo-random spreading code sequence from the mixed echo signal sequence to generate a separate optical contribution point cloud.

[0064] S5.1 The hybrid echo signal sequence and the quasi-orthogonal pseudo-random spreading code sequence are matched in a chaotic synchronous resonance in the multi-scale time-frequency domain to generate a lock signal.

[0065] Furthermore, multi-scale time-frequency analysis is performed on the mixed echo signal sequence. For example, wavelet transform is used to generate its time-frequency energy distribution map at different time and frequency resolutions. Each quasi-orthogonal pseudo-random spreading code sequence is also represented in the same multi-scale time-frequency manner. The matching process is achieved by calculating the cross-correlation coefficient or coherence measure between the time-frequency distribution of the mixed echo signal sequence and the time-frequency distribution of each quasi-orthogonal pseudo-random spreading code sequence. When the time-frequency pattern of a certain quasi-orthogonal pseudo-random spreading code sequence is highly correlated with the time-frequency characteristics of the mixed echo signal sequence at a specific scale, it indicates that the signal component corresponding to the code sequence exists in the mixed signal. At this time, a highly correlated locking signal is generated. The locking signal characterizes the occurrence time, duration, and dominant time-frequency region of the code sequence component in the mixed signal.

[0066] Specifically, the complex time-domain signal separation problem is transformed into a matching problem of finding known pattern features in a multi-scale time-frequency domain. This approach utilizes the inherent broadband characteristics and time-frequency structure of chaotic signals, employing chaotic synchronous resonance matching. Quasi-orthogonal pseudo-random spreading code sequences themselves possess characteristics similar to chaotic signals—broadband, noise-like, and sensitive to initial conditions. The mixed echo signal sequence is a linear mixture of multiple such sequences with noise, laser pulse echoes, etc. In the multi-scale time-frequency domain, the energy of each code sequence is dispersed across a wide frequency band, but at specific time segments and frequency band scales, its own time-frequency structure exhibits identifiable patterns. For example, a code sequence may exhibit high-frequency oscillation dominance in a short period and low-frequency fluctuation dominance in another. The time-frequency distribution of the mixed echo signal sequence can be seen as the result of these patterns superimposed with different intensities. The similarity of local time-frequency structures is sought across multiple time and frequency scales. When a match is successful, the locking signal not only confirms the existence of the code sequence but also indicates its active region in the time-frequency domain.

[0067] S5.2 The locking signal is separated into chaotic attractor signal components corresponding to each code sequence through a self-organizing resonant network.

[0068] Furthermore, the lock signal is used to separate the chaotic attractor signal components corresponding to each code sequence through a self-organizing resonant network. The self-organizing resonant network consists of a set of parallel resonators or dynamic filters. Each resonator is pre-tuned to a specific dynamic mode or attractor structure corresponding to a quasi-orthogonal pseudo-random spreading code sequence. The lock signal is used as an external driving input to the self-organizing resonant network. When the lock signal indicates that a certain code sequence component is active, the resonator in the network tuned to the dynamic mode of that code sequence is activated and resonates. This resonance is strengthened through feedback and adaptive coupling mechanisms, suppressing the activity of other resonators in the network. Each activated resonator outputs a time-varying signal, which corresponds to the chaotic attractor signal component of the quasi-orthogonal pseudo-random spreading code sequence tuned to that resonator. The chaotic attractor signal component is essentially a time-varying estimate of the component belonging to that code sequence in the mixed echo signal sequence.

[0069] Specifically, it mimics the principles of resonance and lateral inhibition in biological neural systems to achieve signal separation, providing a dynamic and adaptive signal extraction mechanism. Each harmonic oscillator is designed as a simple dynamic system capable of reproducing or tracking the dynamic behavior of the attractor of the corresponding code sequence. The locking signal provides clues as to when and where to look for a certain component, while the resonant network is responsible for capturing and tracking the dynamic evolution of the component. For example, the locking signal may indicate the presence of a component of code sequence A within a certain time window. The harmonic oscillator corresponding to code sequence A is activated and begins to oscillate. Its oscillation adaptively adjusts its parameters to synchronize its output with the part of the input signal belonging to code sequence A. At the same time, the lateral inhibition connection of the network reduces the response to components of code sequences B and C. The dynamic tracking mechanism can handle situations where the signal amplitude changes over time and waveform distortion is caused by multipath effects, because the harmonic oscillator tracks the dynamic structure of the attractor, rather than a fixed waveform template, thus extracting weak code sequence components contaminated by noise and interference from the mixed signal.

[0070] S5.3 The chaotic attractor signal components are stabilized by Lyapunov exponential stabilization to restore them to the independent optical signal components of each vehicle lamp unit.

[0071] Furthermore, the chaotic attractor signal components undergo Lyapunov exponent stabilization to recover independent optical signal components for each headlight unit. The Lyapunov exponent quantifies the sensitivity of the chaotic system's trajectory to initial conditions, i.e., the divergence rate between adjacent trajectories. The chaotic attractor signal components may exhibit trajectory instability or deviation due to noise interference or initial estimation errors. Stabilization involves calculating or estimating the maximum Lyapunov exponent of the chaotic attractor signal components and applying appropriate control strategies, such as feedback linearization or perturbation control, to suppress this positive exponent. This stabilizes the dynamic behavior of the chaotic attractor signal components onto a trajectory consistent with the ideal chaotic attractor of its corresponding quasi-orthogonal pseudo-random spreading code sequence. After stabilization, the time-domain waveform of the chaotic attractor signal components becomes regular and predictable, and its envelope or average power directly reflects the waveform of the original headlight unit modulation control command, thus recovering independent optical signal components proportional to the modulated luminous intensity of each headlight unit.

[0072] Specifically, Lyapunov exponent stabilization quantifies the degree to which the observed chaotic attractor signal component deviates from its ideal trajectory by obtaining the Lyapunov exponent. A small control force opposite to the direction of deviation is applied to pull the trajectory back to the ideal attractor. For example, the chaotic attractor signal component corresponding to the left near beam lamp separated from the mixed signal may have a waveform deviation from the chaotic attractor generated by the ideal left near beam lamp code sequence due to multipath interference. Stabilization analyzes the local dynamics of this component, estimates its Lyapunov exponent, and applies control to synchronize its trajectory evolution with the chaotic attractor of the ideal left near beam lamp code sequence. Once synchronized, the dynamics of this component are completely determined by the ideal code sequence, and its amplitude evolution uniquely corresponds to the original light intensity modulation information. This effectively suppresses trajectory distortion introduced by propagation channel distortion and noise, thereby more accurately recovering the original, deterministic modulation information from seemingly random and unstable chaotic observation signals, improving the accuracy and reliability of independent optical signal component recovery.

[0073] S5.4 The independent optical signal components are tensor-fused with the spatial coordinates of the corresponding detection points to form a separate optical contribution point cloud.

[0074] Furthermore, the independent optical signal components are fused with the spatial coordinates of the corresponding detection points using tensors to form a separate optical contribution point cloud. Each detection point has three-dimensional spatial coordinates determined by the lidar scanning control command. For each detection point, the independent optical signal components of each headlight unit recovered from the mixed echo signal sequence are extracted within the corresponding time window, and their characteristic quantities representing light intensity information are obtained, such as the average power within the time window or the integral intensity within a specific symbol period. The characteristic quantities of the independent optical signal components of each headlight unit, together with the spatial coordinates of the detection point, constitute a multi-dimensional data vector. These data vectors of all detection points are organized according to spatial coordinates to form a multi-dimensional tensor data structure. The different dimensions of this tensor correspond to the spatial coordinates and different headlight unit identifiers, respectively. The resulting separate optical contribution point cloud is essentially a dataset, in which each data point not only contains spatial location information but also contains the light intensity information independently contributed by each headlight unit at that location, thereby decoupling the light contributions from different light sources in space.

[0075] Specifically, a multi-dimensional data structure was constructed that tightly couples the light source identity information with spatial geometric information, realizing the mapping from mixed measurement data to the independent spatial light field distribution of each light source. Each spatial point is expanded into a superpoint. In addition to three-dimensional coordinates, this superpoint also carries a vector. Each element in the vector corresponds to the light contribution of a specific vehicle lamp unit at that point. For example, for a certain point on the road, the record of that point in the separated light contribution point cloud not only includes its XYZ coordinates, but also a list of values, representing the light intensity generated by the left low beam, right low beam, high beam, etc. at that point. The data structure completely preserves the most critical information correlation between who and where the light was generated and how much light was generated in the light field measurement. By integrating the multi-light source illumination scenes that are mixed together in the physical world, it is clearly decomposed into the superposition of multiple single-light source illumination scenes at the data level.

[0076] S6. Based on the split optical contribution point cloud, a comprehensive three-dimensional light field distribution is generated through spatial interpolation and superposition reconstruction.

[0077] S6.1 The separated light contribution point cloud is used to generate an independent three-dimensional light field distribution for each vehicle lamp unit through the Kriging space interpolation algorithm.

[0078] Furthermore, the separate light contribution point cloud is used to generate independent three-dimensional light field distributions for each headlight unit through the Kriging spatial interpolation algorithm. For each headlight unit, the independent light signal components belonging to that headlight unit and their corresponding spatial coordinates are extracted from the separate light contribution point cloud, forming a set of spatial light intensity sampling points for that light source. Based on these sparse and potentially non-uniformly distributed sampling points, the Kriging spatial interpolation algorithm constructs a variogram model reflecting the correlation of spatial light intensity to perform optimal unbiased estimation for each grid point on the target three-dimensional spatial grid. The interpolation process considers not only the distance between the sampling point and the point to be estimated, but also the spatial structural relationship between the sampling points, thereby generating a continuous three-dimensional scalar field that matches the measured value at the sampling point and smoothly transitions in the unsampled area, i.e., the independent three-dimensional light field distribution of that headlight unit. This process is repeated for each headlight unit to obtain the independent three-dimensional light field distribution corresponding to each light source.

[0079] Specifically, spatial interpolation techniques from geostatistics are used to address the problem of high-precision reconstruction under sparse and non-uniform light field sampling. The measured light field is treated as a regional variable, and its spatial variation is modeled using a variogram. For example, in front of a road, the spatial continuity of headlight intensity is relatively strong, and the variogram shows high correlation within short distances. However, when encountering the edge of an obstacle, the light intensity may change abruptly, and the variogram reflects spatial discontinuity. The Kriging algorithm utilizes this model. During interpolation, it assigns optimal weights to each sampling point based on the spatial configuration (distance, direction) of the surrounding sampling points and the variogram model, rather than relying solely on the inverse of distance. This fully utilizes the prior spatial correlation of sampling points, providing statistically optimal estimates even in areas with very sparse sampling points. It can also provide the variance of the estimation error, which is very useful for assessing the confidence of the reconstructed light field. Compared to traditional methods, it is more adaptable to non-uniform data generated by intelligent detection, generating independent three-dimensional light field distributions for each headlight unit that conform to the physical laws of light propagation.

[0080] S6.2 The independent three-dimensional light field distributions of each vehicle lamp unit are algebraically superimposed on a three-dimensional spatial grid to generate an algebraic superposition result.

[0081] Furthermore, the independent three-dimensional light field distributions of each headlight unit are algebraically superimposed on a three-dimensional spatial grid to define a unified three-dimensional spatial grid whose range and resolution cover the independent three-dimensional light field distributions of all headlight units. The independent three-dimensional light field distributions of each headlight unit are resampled or mapped onto this unified grid to ensure that each grid point has a corresponding light intensity value in each distribution. Algebraic superposition means that at each grid point, the light intensity values ​​of all headlight units at that point are arithmetically added together. This addition process is linear, reflecting the physical principle of the additivity of light intensity when it propagates in space. After performing this operation on all grid points, the result generated is a new three-dimensional scalar field.

[0082] Specifically, the complex light field reconstruction is decomposed into a two-step strategy of separation followed by synthesis. Synthesis is achieved through linear superposition. By utilizing coding modulation and signal separation techniques, the independent three-dimensional light field distribution of each vehicle headlight unit is obtained at the data level, a distribution unaffected by other light sources. The subsequent algebraic superposition is a simple synthesis based on the known pure components. For example, the independent three-dimensional light field distribution of the left low beam headlight shows its light spot distribution on the left side of the road, while the right low beam headlight shows its distribution on the right side. Algebraic superposition adds these two distributions point by point on the spatial grid to obtain the overall distribution of the low beam light pattern. This avoids the serious ill-conditioned problem encountered when directly estimating the total light field, because the light field of a single light source after separation usually has a simpler spatial pattern and is easier to accurately reconstruct from sparse sampling. Linear superposition is the basic physical principle of light intensity superposition, ensuring the correct physical meaning of the synthesized result. Separate reconstruction allows for independent analysis and adjustment of the contribution of each light source.

[0083] S6.3. Based on the algebraic superposition result, a preliminary three-dimensional light field distribution is generated after Gaussian smoothing filtering.

[0084] Furthermore, the algebraic superposition result is processed by Gaussian smoothing filtering to generate a preliminary three-dimensional light field distribution. Gaussian smoothing filtering is achieved by applying a three-dimensional Gaussian convolution kernel to the three-dimensional spatial grid. The standard deviation of the convolution kernel in each dimension determines the degree of smoothing. During the filtering process, the value of each grid point in the algebraic superposition result is replaced by a weighted average of the values ​​of its neighboring grid points. The weights are determined by a Gaussian function, with higher weights for neighboring points closer to the center point. Gaussian smoothing filtering can suppress local estimation noise that may be introduced by Kriging interpolation, block artifacts caused by discrete sampling, and non-physical discontinuities that may occur at the boundaries of algebraic superposition. After smoothing, the generated preliminary three-dimensional light field distribution is more spatially smooth and continuous, and closer to the smooth transition characteristics that a real physical light field should have due to atmospheric scattering and surface reflection diffusion.

[0085] Specifically, smoothing techniques from image processing are introduced into the post-processing of 3D light fields to conform to the prior spatial continuity of physical light fields. This compensates for the non-physical details that may arise from pure mathematical interpolation and superposition. Kriging interpolation is based on statistical optimality, but may produce unnatural fluctuations in some areas due to insufficient data. Algebraic superposition is an ideal linear addition that does not consider the natural blurring effect caused by scattering and diffraction of light during propagation. A reasonable physical assumption is introduced: the light field intensity distribution in real space will not change drastically at the microscopic scale. Due to the wave nature of light and the scattering of the propagation medium, changes in light intensity are usually gradual. Gaussian smoothing is essentially a low-pass filter. For example, near the cutoff line between light and dark, the ideal result of algebraic superposition might be a sharp brightness boundary line. However, in reality, due to factors such as the light distribution lens of the lamp and atmospheric scattering, the cutoff line is a gradient band with a certain width. Gaussian smoothing can simulate the natural blurring effect, making the generated preliminary 3D light field distribution more consistent with actual observations.

[0086] S6.4 The preliminary three-dimensional light field distribution and vehicle attitude parameters are subjected to coordinate transformation correction. The result after coordinate transformation correction is then processed by light intensity normalization to generate a comprehensive three-dimensional light field distribution. Furthermore, the preliminary three-dimensional light field distribution is corrected using coordinate transformation with vehicle attitude parameters, including the pitch and yaw angles acquired in real-time from the inertial measurement unit. Coordinate transformation aims to convert the preliminary three-dimensional light field distribution from the instantaneous coordinate system upon which it was reconstructed, which may change with vehicle attitude, to a standard coordinate system fixed to the vehicle body, or to a stable reference system based on the horizontal plane and the vehicle's longitudinal axis. This correction is achieved through a three-dimensional rotation matrix, which rotates the coordinates of each spatial point in the preliminary three-dimensional light field distribution accordingly. The light intensity value at each spatial point remains unchanged under this purely geometric coordinate transformation. The result after coordinate transformation correction is then subjected to light intensity normalization. Light intensity normalization aims to eliminate the uncertainty in the absolute light intensity scale caused by factors such as measurement distance, atmospheric attenuation, or detector gain, mapping the light intensity value to a standardized range or calibrating it based on a known reference brightness. The final generated three-dimensional scalar field, with normalized light intensity and stabilized coordinates, is the comprehensive three-dimensional light field distribution. The comprehensive three-dimensional light field distribution characterizes the relative light field distribution generated by the vehicle lighting system in three-dimensional space under standard vehicle posture and normalized illumination intensity.

[0087] Specifically, the reconstructed light field is processed using vehicle attitude parameters to remove jitter and anchor the light field distribution to the vehicle itself. For example, when a vehicle is going uphill, the headlights actually shine diagonally upwards, and the initial reconstructed three-dimensional light field distribution will be too high. By combining the pitch angle with reverse rotation correction, this distribution is corrected back to the illumination position that should be present when the vehicle is parked horizontally. The light intensity normalization process solves the problem that absolute brightness measurement is greatly affected by the environment, shifting the focus to the relative spatial distribution of the light field. This is consistent with the characteristic that regulations mainly constrain light patterns rather than absolute illuminance, making the comprehensive three-dimensional light field distribution a fingerprint of lighting performance. It eliminates the interference caused by changes in vehicle dynamics and measurement environment, and can be compared fairly and accurately with pre-stored regulatory light pattern parameters defined under standard conditions.

[0088] S7. Compare the three-dimensional light field distribution with the pre-stored regulatory light pattern parameters and dynamic target space information, and output the vehicle lighting adaptive control command.

[0089] S7.1 The three-dimensional light field distribution is projected onto a standard test screen to generate actual light pattern parameters.

[0090] Furthermore, the three-dimensional light field distribution is projected onto a standard test screen to generate actual light pattern parameters. The standard test screen is a virtual or conceptual plane, typically located at a specific distance in front of the vehicle, such as the distance specified in regulatory testing. The projection process involves extracting the light intensity values ​​of all spatial points located within the light cone formed by the vehicle's headlight center pointing towards this virtual plane from the three-dimensional light field distribution, and mapping the light intensity of these three-dimensional spatial points onto two-dimensional coordinate points on the virtual plane according to perspective projection relationships. For each sampling point coordinate on the virtual plane, the equivalent illumination intensity of that point on the virtual plane is calculated by integrating all three-dimensional light intensity values ​​projected to its vicinity, thereby generating a two-dimensional illumination intensity distribution map, i.e., the actual light pattern parameters.

[0091] Specifically, by projecting onto a two-dimensional test screen defined by regulations, the data is converted into a two-dimensional form that can be directly compared point-by-point with standards, bridging the gap between complex three-dimensional perception results and traditional two-dimensional regulatory standards. The comprehensive three-dimensional light field distribution is a data volume containing complete spatial information, simulating the imaging principle of real light on the test screen. It involves the integration process from a three-dimensional volume to a two-dimensional plane. For example, a light source at a high position may illuminate a lower position on the screen, determined by geometric optics. The projection algorithm tracks the light rays from the headlights to each point on the screen and accumulates the light intensity contributions along the light path in the three-dimensional light field distribution to obtain the total illuminance at that point on the screen. The actual light pattern parameters include not only direct light but also light contributions indirectly reaching the screen after being scattered by the environment. This provides a more comprehensive digital simulation of the real lighting effect, generating actual light pattern parameters that can be directly used for compliance comparison.

[0092] S7.2 Calculate the difference between the actual light pattern parameters and the pre-stored regulatory light pattern parameters to generate a compliance deviation matrix.

[0093] Furthermore, the difference between the actual light pattern parameters and the pre-stored regulatory light pattern parameters is calculated to generate a compliance deviation matrix. The calculation process is performed on a regular grid on a two-dimensional virtual detection plane. For each coordinate point on the grid, the illumination intensity value of the actual light pattern parameter at that point is read, along with the corresponding standard illumination intensity threshold from the pre-stored regulatory light pattern parameters. The threshold is typically a range; the upper limit, lower limit, or median value can be used during the difference calculation, depending on the compliance judgment logic. Each element of the compliance deviation matrix... , >0 indicates that the actual illumination exceeds the standard limit. <0 indicates below the standard lower limit. The generated compliance deviation matrix is ​​a two-dimensional numerical matrix, whose row and column indices correspond to the sampling grid of the virtual detection plane. Each value in the matrix quantifies the degree and direction of compliance deviation at the corresponding spatial point.

[0094] Specifically, the subjective, overall judgment of light pattern compliance is transformed into an objective, quantified spatial point-by-point deviation data matrix, enabling data-driven precise control decisions. The compliance status of the entire light illumination pattern is decomposed into the local compliance status of thousands of independent grid points. For example, the compliance deviation matrix can clearly show whether the brightness at the inflection point of the left light-dark cutoff line slightly exceeds the upper limit, or whether individual points in the right glare area severely exceed the standard. Fine-grained spatial deviation information is the direct basis for subsequent targeted and adaptive light adjustments. As a structured data representation, the compliance deviation matrix can be directly input into the control algorithm to drive the headlight unit to perform pixel-level or zone-level brightness adjustments to minimize the deviation values ​​in the matrix, thereby achieving truly refined adaptive light control.

[0095] The expression for the compliance deviation matrix is: ; in, The first in the compliance deviation matrix row and number Column elements, To the coordinate points of the virtual detection plane At that location, the actual illumination intensity value obtained by projecting the three-dimensional light field distribution. To be at the same coordinate point The standard illuminance threshold specified by regulations for light pattern parameters is pre-stored. For the first The horizontal coordinates of each sampling point. For the first The vertical coordinates of each sampling point. This refers to the sampling point number in the horizontal direction. This refers to the sampling point number in the vertical direction.

[0096] S7.3. Spatial intersection detection is performed between the three-dimensional light field distribution and the dynamic target spatial information to generate the target-affected... The system compares the light intensity of the illuminated area with the glare threshold to generate a glare risk assessment result.

[0097] Furthermore, spatial intersection detection is performed between the 3D light field distribution and the dynamic target spatial information to generate the target's illuminated area. The dynamic target spatial information describes the position, outline, and trajectory of the moving target in front. Spatial intersection detection is achieved by calculating the geometric intersection between the spatial range of the 3D light field distribution and the 3D occupied space (e.g., represented by a bounding box or point cloud) of each dynamic target described in the dynamic target spatial information. All spatial points within the intersection volume constitute the target's illuminated area. For the target's illuminated area, the illumination intensity values ​​of all points within the area are extracted from the 3D light field distribution and compared with a preset glare threshold. The comparison can check whether the light intensity of any point exceeds the threshold, calculate the proportion of points in the area whose light intensity exceeds the threshold, or calculate the average light intensity in the area. Based on the comparison results, a glare risk assessment result is generated, which can be a Boolean value indicating the presence of glare risk, or a risk level or risk score.

[0098] Specifically, by combining static light field distribution analysis with dynamic target information, dynamic anti-glare risk assessment for traffic participants is achieved. This surpasses the traditional approach of merely conducting static compliance checks on the lighting pattern itself. Regulatory light pattern parameters are static standards, while real traffic scenarios are dynamic. It no longer only concerns whether the pattern of the light on a fixed screen is compliant, but also whether the light will cause uncomfortable glare to other road users on the actual road. For example, even if the cutoff line of the low beam headlights fully complies with regulations, if there is a motorcycle ahead below the ramp, its driver's eyes may happen to be in the bright area of ​​the light spot, which will cause glare. By spatially intersecting the three-dimensional light field distribution with the dynamic target spatial information, it can be determined whether the light has shone on the sensitive areas of other traffic participants (such as the estimated eye position). The glare risk assessment results quantify the risk, elevating the light performance assessment from a fixed pattern to a dynamic scenario.

[0099] S7.4 Input the compliance deviation matrix and glare risk assessment results into the fuzzy logic controller, and output... Vehicle lighting adaptive control command.

[0100] Furthermore, the statistical characteristics of the compliance deviation matrix (such as the maximum positive deviation, the area of ​​the negative deviation region, and the overall root mean square deviation) and the glare risk assessment results (such as risk level and the location of the risk target) are transformed into fuzzy linguistic variables through a membership function. Based on a pre-defined fuzzy rule base (e.g., if there is a significant positive deviation in the cutoff line area, reduce the brightness of the lights in the corresponding area; if there is a high glare risk target ahead, reduce the brightness of the lights illuminating that target), fuzzy inference is performed. The inference process may use Mamdani or Sugeno fuzzy inference methods to obtain fuzzy outputs on how each headlight unit or each headlight zone should be adjusted. Finally, through a defuzzification process, these fuzzy outputs are converted into specific and clear vehicle headlight adaptive control commands. These commands can be instructions to adjust the brightness of a specific headlight unit or instructions to control the on / off state or brightness of specific pixels in the matrix headlights.

[0101] Specifically, adaptive lighting control needs to balance multiple potentially conflicting objectives, such as meeting regulations, avoiding glare, and ensuring its own lighting needs. Furthermore, the control logic is difficult to describe with precise mathematical formulas. Therefore, it encodes human engineers' dimming experience into fuzzy rules. The compliance deviation matrix provides precise but local information on where the lighting is too bright or too dark, while the glare risk assessment results provide global safety constraints on whether to allow pedestrians to pass. The fuzzy logic controller can integrate this information to make trade-off decisions. For example, when a pedestrian is detected on the right with a glare risk, but the right-side road edge lighting is insufficient, the fuzzy rules might drive the controller to issue an instruction: slightly reduce the brightness of a few pixels directly illuminating the upper part of the pedestrian's body to reduce glare, while maintaining or slightly increasing the brightness of pixels illuminating the road surface near the pedestrian's feet to ensure lighting safety. This rule-based and approximate reasoning decision-making approach is insensitive to sensor noise and model uncertainty, can handle complex, nonlinear control tasks, and its rules are easy to understand and adjust. This makes it suitable for decision-making scenarios requiring high safety and intuitiveness, such as vehicle headlight control, generating adaptive vehicle lighting control instructions.

[0102] This embodiment also provides a computer device applicable to the method for detecting the illumination range of motor vehicle lights based on lidar, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the method for detecting the illumination range of motor vehicle lights based on lidar as proposed in the above embodiment.

[0103] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0104] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the method for detecting the illumination range of motor vehicle lights based on lidar as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0105] In summary, this invention combines directional scanning with lidar and concurrent coding modulation of the vehicle lighting unit. It uses quasi-orthogonal pseudo-random spreading code sequences to uniquely identify each vehicle lighting unit, separates the independent light contribution of each light source from the mixed echo signal, constructs a separate light contribution point cloud, and then generates a comprehensive three-dimensional light field distribution through spatial interpolation and superposition reconstruction. This distribution is compared with regulatory light pattern parameters and dynamic target information to output adaptive control commands. This achieves precise decoupling and real-time evaluation of the spatial energy distribution of each independent light source in complex multi-light source scenarios, providing vehicle lighting units with light source-level fine control capabilities.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting the illumination range of motor vehicle headlights based on lidar, characterized in that: include, Initial environmental perception baseline information is obtained through the vehicle's sensing unit; The initial environmental perception baseline information is input into the reinforcement learning decision engine, and a set of high-value detection points is output based on the information gain maximization strategy. The lidar is controlled to perform directional scanning based on the high-value detection point set, and the vehicle lamp unit under test is controlled to perform concurrent modulation and emission according to the pre-allocated quasi-orthogonal pseudo-random spreading code sequence, and the mixed echo signal sequence of the coding characteristics of each vehicle lamp unit is collected. Based on the principle of code coherent resonance, independent optical signal components corresponding to quasi-orthogonal pseudo-random spreading code sequences are separated from the mixed echo signal sequence to generate a separate optical contribution point cloud; A comprehensive three-dimensional light field distribution is generated by spatial interpolation and superposition reconstruction based on the separated light contribution point cloud. The three-dimensional light field distribution is compared with the pre-stored regulatory light pattern parameters and dynamic target spatial information to output vehicle lighting adaptive control commands.

2. The method for detecting the illumination range of motor vehicle headlights based on lidar as described in claim 1, characterized in that: The initial environmental perception baseline is obtained through the vehicle's sensing unit, including the following steps: The vehicle's front area is scanned by LiDAR to generate initial point cloud data with three-dimensional coordinates and reflection intensity. Visible light image data of the front environment is collected by the vehicle's onboard camera. The relative speed and azimuth data of moving targets ahead are detected by millimeter-wave radar, and the pitch and yaw angle data of the vehicle are measured in real time using an inertial measurement unit. The initial point cloud data, visible light image data, relative velocity and azimuth data, and pitch and yaw angle data are input into the multi-sensor fusion algorithm to output the initial environmental perception reference information.

3. The method for detecting the illumination range of motor vehicle headlights based on lidar as described in claim 2, characterized in that: The initial environmental perception baseline is input into the reinforcement learning decision engine, which outputs a set of high-value detection points based on the information gain maximization strategy. This includes the following steps: The initial environmental perception benchmark is used to extract multi-dimensional spatial light field uncertainty features through a spatiotemporal attention encoder to form a state representation vector; The state representation vector is input into the light field exploration network of the reinforcement learning decision engine. Based on the information gain maximization strategy, Monte Carlo tree search is performed in the potential probe point space to obtain the expected information gain value. Potential probe points whose expected information gain values ​​conform to the dynamic Pareto front are selected to form a candidate probe point set. Spatial coverage optimization based on the Voronoi diagram is performed on the candidate probe point set to form a high-value probe point set.

4. The method for detecting the illumination range of motor vehicle headlights based on lidar as described in claim 3, characterized in that: The lidar is controlled to perform directional scanning based on a set of high-value detection points, and the headlight unit under test is controlled to emit light concurrently according to a pre-assigned quasi-orthogonal pseudo-random spreading code sequence, including the following steps: The high-value detection point set is converted into lidar scanning control commands, which drive the lidar to perform directional scanning. The quasi-orthogonal pseudo-random spreading code sequence is converted into a vehicle lamp unit modulation control command. The vehicle lamp unit modulation control command drives the vehicle lamp unit under test to perform concurrent modulation and emission. The directional scanning and concurrent modulation and emission are executed in time-aligned. The lidar receives reflected signals from the detection area, forming a mixed echo signal sequence.

5. The method for detecting the illumination range of motor vehicle headlights based on lidar as described in claim 4, characterized in that: The process of acquiring the mixed echo signal sequence of the coding features of each vehicle headlight unit includes the following steps: The concurrently modulated light signal is reflected by the object surface within the detection area, and the reflected light signal and the ambient background light form an optical superposition during spatial propagation. The optical superposition signal is received by the photodetector of the lidar and converted into an electrical signal. The electrical signal is amplified by a transimpedance amplifier to form an analog signal. The analog signal is digitized by a high-speed analog-to-digital converter at a sampling rate. The digitized signal is then segmented and organized according to the detection points to form a mixed echo signal sequence.

6. The method for detecting the illumination range of motor vehicle headlights based on lidar as described in claim 5, characterized in that: Based on the principle of code pattern coherence resonance, independent optical signal components corresponding to quasi-orthogonal pseudo-random spreading code sequences are separated from the mixed echo signal sequence to generate a separated optical contribution point cloud, including the following steps: The hybrid echo signal sequence and the quasi-orthogonal pseudo-random spreading code sequence are matched in a chaotic synchronous resonance in the multi-scale time-frequency domain to generate a lock signal. The locking signal is separated into the chaotic attractor signal components corresponding to each code sequence through a self-organizing resonant network. The chaotic attractor signal components are stabilized by Lyapunov exponential stabilization to restore the independent optical signal components of each vehicle lamp unit. The independent optical signal components are tensor-fused with the spatial coordinates of the corresponding detection points to form a separate optical contribution point cloud.

7. The method for detecting the illumination range of motor vehicle headlights based on lidar as described in claim 6, characterized in that: Based on the separated optical contribution point cloud, a comprehensive three-dimensional light field distribution is generated through spatial interpolation and superposition reconstruction, including the following steps: The separated light contribution point cloud is used to generate an independent three-dimensional light field distribution for each headlight unit through the Kriging space interpolation algorithm. The independent three-dimensional light field distributions of each vehicle headlight unit are algebraically superimposed on a three-dimensional spatial grid to generate an algebraic superposition result; Based on the algebraic superposition result, a preliminary three-dimensional light field distribution is generated after Gaussian smoothing filtering. The preliminary three-dimensional light field distribution is corrected by coordinate transformation with the vehicle attitude parameters. The result after coordinate transformation correction is normalized by light intensity to generate a comprehensive three-dimensional light field distribution.

8. The method for detecting the illumination range of motor vehicle headlights based on lidar as described in claim 7, characterized in that, The three-dimensional light field distribution is compared with pre-stored regulatory light pattern parameters and dynamic target spatial information to output vehicle lighting adaptive control commands, including the following steps: The three-dimensional light field distribution is projected onto a standard test screen to generate actual light pattern parameters; The difference between the actual beam pattern parameters and the pre-stored regulatory beam pattern parameters is calculated to generate a compliance deviation matrix; Spatial intersection detection is performed between the three-dimensional light field distribution and the dynamic target spatial information to generate the target illuminated area. The glare risk assessment results are generated by comparing the light intensity of the target illuminated area with the glare threshold. The compliance deviation matrix and glare risk assessment results are input into the fuzzy logic controller, which outputs vehicle lighting adaptive control commands.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for detecting the illumination range of motor vehicle lights based on lidar as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for detecting the illumination range of motor vehicle lights based on lidar as described in any one of claims 1 to 8.

Images