Sightseeing vehicle anti-collision control method and system based on multi-mode radar

Abstract

The invention relates to the field of automatic driving vehicle control, in particular to a sightseeing vehicle anti-collision control method and system based on a multi-mode radar. Comprising the following steps: acquiring a multi-modal radar original signal, and performing timestamp alignment and noise reduction processing to generate standardized fusion data; performing multi-target classification based on a convolutional neural network, and outputting a target recognition result in combination with Doppler frequency shift compensation and distance calculation; generating a hierarchical response strategy through dual-threshold comparison and environment interference dynamic weight adjustment; the strategy is converted into a PWM control signal, and a control instruction is generated through relay driving adaptation and hardware compatibility detection; a relay is driven in real time, motor response is collected, and execution effect evaluation and time sequence marking are completed; and brake release and system parameter synchronization are realized through PID closed-loop control. The problems of insufficient multi-target identification precision, poor environmental interference adaptability and response delay are solved, and the active safety performance of the sightseeing vehicle is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automatic driving vehicle control, and in particular to a sightseeing vehicle anti-collision control method and system based on multi-modal radar. BACKGROUND

[0002] In the field of automatic driving vehicle control, the existing scheme usually relies on radar sensors, relay control modules, signal processing units, etc., and has limitations such as untimely dynamic response, insufficient safety and stability, and poor adaptability to complex environments.

[0003] The existing method usually realizes vehicle anti-collision through preset control logic and simple signal processing, which is prone to misjudgment and control failure in complex environments, and is difficult to meet the stable implementation of the sightseeing vehicle anti-collision control system.

[0004] In view of the joint processing of how to implement the record based on multi-modal radar, through the relay execution control module and the multi-modal radar signal fusion processing module to realize the dynamic response and safety and stability of the sightseeing vehicle anti-collision control system in complex environments, the existing technology generally has the shortcomings of low judgment accuracy and slow response speed in the aspects of signal fusion, state determination, control adjustment, etc., and it is difficult to form a consistent process of closed-loop linkage between the relay execution control and the multi-modal radar signal fusion processing in complex environments, resulting in insufficient dynamic response and safety and stability of the sightseeing vehicle anti-collision system. SUMMARY

[0005] The present application provides a sightseeing vehicle anti-collision control method and system based on multi-modal radar to solve the problem of how to implement the record based on multi-modal radar, through the relay execution control module and the multi-modal radar signal fusion processing module to realize the dynamic response and safety and stability of the sightseeing vehicle anti-collision control system in complex environments.

[0006] In order to solve the above technical problems, the present application provides a sightseeing vehicle anti-collision control method based on multi-modal radar, comprising:

[0007] Collect 24G microwave radar and ultrasonic radar original signals, perform timestamp alignment and noise reduction, and generate standardized fusion data; the original signals are accessed to the system through a multi-channel data acquisition interface, and the system is based on a preset sampling frequency and a time synchronization mechanism; a hardware trigger and a software time correction combination are used to capture the signal sampling time in real time, and a timestamp matching algorithm is used to eliminate abnormal time sequence data; an adaptive filter and wavelet transform technology are combined to suppress high-frequency noise and low-frequency drift components respectively; noise peaks are identified through a dynamic threshold adjustment mechanism, and abnormal signal segments are interpolated and corrected;

[0008] Based on the fusion data, a convolutional neural network is used to realize multi-target classification, Doppler shift compensation and distance calculation, and output target recognition results;

[0009] Performing double-threshold comparison and environmental interference dynamic weight adjustment on the target recognition result, completing timing conflict detection, and generating a hierarchical response strategy;

[0010] Converting into PWM control signals according to the hierarchical response strategy, extracting voltage amplitude to complete relay driving signal adaptation and hardware compatibility detection, and generating control instructions;

[0011] Receiving control instructions, driving the relay to act and collecting motor responses in real time, completing execution effect evaluation and timing marking, and generating an execution record with a timestamp;

[0012] Analyzing the execution record to determine the brake release condition, generating adjustment instructions based on PID control, and feeding back configuration closed-loop parameters to achieve system parameter synchronization and closed-loop control; the expression of the feedback configuration closed-loop parameter includes:

[0013]

[0014] wherein, is the optimized loop gain; is the regularization coefficient; is the gradient of the synchronous voltage to the gain; is the state determination gradient; is the regularization coefficient; is the L2 norm; is the L1 norm; is the specific G value that takes the minimum value; is the original variable of the loop gain to be optimized.

[0015] Further, based on the fusion data, a multi-target classification is realized through a convolutional neural network, Doppler shift compensation and distance calculation are performed, and a target recognition result is output, specifically including:

[0016] The fusion data is connected to a convolutional neural network target detection sub-module through a high-speed bus interface, and the convolutional neural network is based on a pre-trained model combined with a multi-layer feature extraction structure to perform multi-scale spatial feature analysis and time series feature capture on the input fusion data;

[0017] The convolutional neural network adopts a multi-layer convolutional layer and a pooling layer alternately stacked, combined with batch normalization and an activation function processing, to extract human motion features and obstacle contour features;

[0018] A multi-class probability distribution is output through a Softmax layer to realize preliminary classification of human bodies, static obstacles and dynamic obstacles;

[0019] A dynamic threshold filtering mechanism is set to eliminate low-confidence targets to prevent false positives;

[0020] The target classification result is recorded in the target identification log for subsequent performance evaluation and model updating.

[0021] Further, based on the fusion data, the multi-target classification is realized through the convolutional neural network, the Doppler frequency shift compensation and the distance calculation are performed, and the process of outputting the target identification result further includes:

[0022] The spatial positioning information and the speed estimation data in the target classification result are analyzed as the basic input of the distance parameter;

[0023] In combination with the Doppler frequency shift compensation algorithm, the frequency offset in the target echo signal is corrected, and the frequency drift caused by the motion of the sightseeing vehicle or the target dynamic is eliminated;

[0024] The Doppler frequency shift compensation algorithm adopts a spectrum analysis method based on fast Fourier transform, and combines real-time speed sensor data to dynamically adjust the compensation coefficient;

[0025] Through the time delay ranging technology, the time difference measurement of the multi-modal radar is combined to calculate the accurate distance value between the target and the sightseeing vehicle;

[0026] The ranging result is sampled and weighted averaged for multiple times, and the abnormal value is eliminated, and the filtering parameter is adjusted according to the environmental noise level.

[0027] Further, the target identification result is subjected to double-threshold comparison and dynamic weight adjustment of environmental interference, specifically including:

[0028] The target distance value is compared with the preset 8-meter and 1.5-meter threshold values one by one;

[0029] When the target distance is less than or equal to 8 meters and the target type is human body, the first-level response condition is determined;

[0030] When the target distance is less than or equal to 1.5 meters and the target type is an obstacle, the second-level response condition is determined;

[0031] A hardware timing trigger mechanism is adopted, and software algorithms are used to filter multiple continuous frames of data, thereby reducing the probability of misjudgment;

[0032] The comparison result is assigned a preliminary response level, the first-level response corresponds to a speed limit warning level, the second-level response corresponds to an emergency braking level, and if neither of them is satisfied, the response level is no response.

[0033] Further, the process of dynamic weight adjustment of environmental interference includes:

[0034] The light intensity, temperature change and radar signal noise level of the current environment are obtained through the environment monitoring module, and the environmental interference coefficient is calculated in combination with historical data and real-time sampling;

[0035] The response level and the environmental interference coefficient are input into a dynamic weight adjustment algorithm based on a fuzzy logic controller design to dynamically adjust the weight parameters of the response instructions;

[0036] A sliding window mechanism is used to smooth the weight changes to prevent instruction fluctuations caused by sudden environmental changes;

[0037] An optimized response instruction is generated according to the adjusted weight parameters, and the instruction content includes the response level, priority identification and execution time window.

[0038] Further, the process of completing the timing conflict detection and generating the hierarchical response strategy specifically includes:

[0039] A timing event sequence is constructed to identify the time overlap, priority conflict and execution resource competition between instructions;

[0040] The time windows of the optimized instructions are sorted to detect whether there is an overlapping interval of instruction execution time;

[0041] If overlap is found, the instruction priority is further analyzed, the instruction with high priority is given execution right, and the low-priority instruction is delayed or canceled;

[0042] Combined with the relay execution state feedback, the current hardware execution capability is confirmed to avoid hardware damage or response failure caused by instruction conflict;

[0043] For the detected abnormal conflict situation, the conflict resolution mechanism is automatically triggered, including instruction rescheduling, priority re-arbitration and temporary response level adjustment.

[0044] Further, the process of converting the hierarchical response strategy into a PWM control signal specifically includes:

[0045] According to the preset PWM signal conversion rule, the discrete response level is mapped to the corresponding PWM duty cycle and frequency parameters;

[0046] A hardware timer and a digital signal generation module are used to work together, the timer accurately controls the PWM period according to the system clock period, and the digital signal generation module adjusts the duty cycle according to the mapping rule;

[0047] The amplitude and frequency of the PWM waveform are monitored in real time, and a closed-loop feedback mechanism is used to fine-tune the waveform to avoid waveform distortion caused by hardware jitter or electromagnetic interference;

[0048] An abnormality detection mechanism is set, when the PWM waveform exceeds the preset amplitude or frequency range, an error flag is triggered and an abnormal event log is recorded.

[0049] Further, the extraction of voltage amplitude completes the relay drive signal adaptation and hardware compatibility detection, and the process of generating control instructions specifically includes:

[0050] The PWM waveform is converted into a voltage signal sample through the analog signal acquisition interface, and digital sampling is completed by using a high-precision analog-to-digital converter.

[0051] Synchronous sampling technology is adopted to ensure that the sampling time is strictly aligned with the PWM signal period.

[0052] The voltage amplitude data obtained by sampling is processed by a digital filter, and a finite impulse response filter is used to eliminate high-frequency noise and transient interference.

[0053] The processed voltage amplitude is converted into the level parameter of the relay drive signal through a voltage amplitude mapping algorithm.

[0054] According to the relay drive voltage threshold, the voltage threshold is dynamically adjusted to ensure the reliability and stability of the relay drive.

[0055] Further, the expression of the system parameter synchronization and closed-loop control based on the analysis of the execution record and the determination of the brake release condition includes:

[0056] A dynamic adjustment mechanism for defining the brake release trigger threshold is defined:

[0057]

[0058] wherein, is the dynamic brake release threshold; , , is the weight coefficient; is the basic threshold; is the energy gradient of the kth execution signal; is the kth environmental disturbance factor; is the total number of environmental disturbance factor categories; is the environmental disturbance intensity index;

[0059] Further, the logic synthesis of state determination is constructed:

[0060]

[0061] wherein, is the brake release state flag; is the activation function; is the normalization coefficient; is the continuous multiplication operator; is the index variable; is the kth a confidence weight of the history execution record; a history execution matching degree; a history execution record number, and an index upper limit;

[0062] based on and a speed recovery parameter, performing a PID control calculation to construct a proportional-integral-derivative term:

[0063]

[0064] wherein, a PID control output instruction; a speed error; a target speed limit value; a current vehicle speed; , , a dynamic PID coefficient; an environmental disturbance coefficient; an integral time interval;

[0065] Further, the safety boundary constraint of the control instruction is:

[0066]

[0067] wherein, a final adjustment instruction; a PID output change gradient; a gradient change threshold; , a safety range of the control instruction; a limiting function; an L2 norm;

[0068] Further, based on , a closed-loop feedback configuration is generated to define a time domain mapping of parameter synchronization:

[0069]

[0070] wherein, a synchronized actuator driving voltage; a time decay weight; a decay coefficient; a current time; a first history execution timestamp; a history execution record number.

[0071] Further, a sightseeing vehicle anti-collision control system based on a multi-modal radar is applied to any one of the above-mentioned methods, comprising:

[0072] An installation and calibration unit is configured to obtain multi-modal radar installation parameters and complete sensor calibration;

[0073] A sector limiting unit is configured to set the horizontal field of view coverage range of the millimeter wave radar;

[0074] A threshold and strategy configuration unit is configured to configure the collision risk judgment threshold and the dynamic arbitration strategy table;

[0075] A data acquisition unit is configured to acquire the ranging and speed data of the millimeter wave radar and the point cloud data of the laser radar;

[0076] A time synchronization unit is configured to align the timestamps of the multi-modal radars and perform spatial coordinate mapping;

[0077] An event judgment unit is configured to generate a collision risk level based on the fused target trajectory sequence;

[0078] An arbitration unit is configured to decide the hierarchical control instruction according to the priority strategy table;

[0079] A speed limiting signal output unit is configured to send a deceleration instruction to a power controller;

[0080] A brake signal output unit is configured to send an emergency brake instruction to a brake actuator;

[0081] An alarm management unit is configured to trigger an audible and visual alarm device;

[0082] A log and parameter updating unit is configured to record the control instruction timestamp and update the strategy table parameters.

[0083] The key innovations of the present application include:

[0084] (1) The system accesses the execution record and performs format analysis and integrity check by taking the execution record of the relay execution control module as the input source, to ensure that the data is complete and conforms to the preset protocol specification.

[0085] (2) The system adopts a multi-threshold judgment mechanism, combines environmental perception parameters and historical execution records, dynamically adjusts the brake release trigger threshold, and determines whether the current execution state meets the conditions for releasing the brake through a state machine model.

[0086] (3) The system generates an adjustment instruction suitable for the current operating environment through proportional, integral and differential calculation based on the speed recovery parameter in the state determination result through the PID controller module, and dynamically adjusts the PID parameters in combination with the environmental disturbance coefficient.

[0087] The main beneficial effects are as follows:

[0088] (1) By ensuring the data integrity and protocol specification of the execution record, the system can effectively receive and process the input signals of the relay execution control module, form a reliable input data link, and provide a stable data foundation for subsequent state judgment and control adjustment.

[0089] (2) The combination of multi-threshold judgment mechanism and state machine model enables the system to accurately determine the brake release condition in complex environments, avoid misjudgment caused by short-term abnormal signals, and ensure the closed-loop linkage between relay execution control and multi-modal radar signal fusion processing.

[0090] (3) The PID controller module dynamically adjusts the parameters to ensure that the generated adjustment instructions adapt to the current environmental changes, enhancing the robustness and response speed of the control, preventing overshoot or under-shoot, and ensuring the continuous, stable and efficient operation of the sightseeing vehicle collision avoidance system. BRIEF DESCRIPTION OF DRAWINGS

[0091] Figure 1 A flowchart of a sightseeing vehicle collision avoidance control method based on a multi-modal radar provided by an embodiment of the present application;

[0092] Figure 2 A structural block diagram of a sightseeing vehicle collision avoidance control system based on a multi-modal radar provided by an embodiment of the present application. DETAILED DESCRIPTION

[0093] Embodiment one: reference Figure 1 is a flowchart of a sightseeing vehicle collision avoidance control method based on a multi-modal radar provided by an embodiment of the present application, which can at least include steps S100-S600:

[0094] S100, collect 24G microwave radar and ultrasonic radar original signals, perform timestamp alignment and noise reduction, and generate standardized fusion data.

[0095] S200, based on the fusion data, realize multi-target classification through a convolutional neural network, perform Doppler shift compensation and distance calculation, and output target recognition results.

[0096] S300, perform double-threshold comparison and dynamic weight adjustment of environmental interference on the target recognition results, complete timing conflict detection, and generate a hierarchical response strategy.

[0097] S400, convert the hierarchical response strategy into a PWM control signal, extract the voltage amplitude to complete relay drive signal adaptation and hardware compatibility detection, and generate a control instruction.

[0098] S500, receive the control instruction, drive the relay to act and collect the motor response in real time, complete the execution effect evaluation and timing marking, and generate an execution record with a timestamp.

[0099] S600, analyze the execution record to determine the brake release condition, generate an adjustment instruction based on PID control, configure a closed-loop parameter for feedback, and achieve system parameter synchronization and closed-loop control.

[0100] Step S100 includes at least steps S110-S130:

[0101] S110, obtain the original signals of the 24G microwave radar and the ultrasonic radar, perform time alignment and noise reduction processing, and obtain preprocessed signals.

[0102] Specifically, the original signals of the 24G microwave radar and the ultrasonic radar are collected as the input source of this step. First, the original signals are accessed to the system through a multi-channel data acquisition interface. The system performs timestamp alignment processing on the original signals of the 24G microwave radar and the ultrasonic radar according to a preset sampling frequency and a time synchronization mechanism, to ensure the synchronization of different radar signals on the time axis. Specifically, a combination of hardware triggering and software time correction is used to capture the signal sampling time in real time, and a timestamp matching algorithm is used to eliminate abnormal timing data, to ensure the timing consistency of the signals. Subsequently, the preprocessing process performs noise reduction processing on the time-aligned signals. Specifically, adaptive filters and wavelet transform techniques are combined to suppress high-frequency noise and low-frequency drift components, respectively, to enhance the effective components of the signals. In the noise reduction process, the system identifies noise peaks through a dynamic threshold adjustment mechanism and performs interpolation correction on abnormal signal segments to prevent signal distortion. The processing status and abnormal information of the preprocessed signals are recorded in the system log for subsequent diagnosis.

[0103] S120, extract the human feature spectrum and obstacle echo intensity from the preprocessed signals, perform frequency domain feature fusion, and generate a fused feature vector.

[0104] Further, the human feature spectrum and the obstacle echo intensity are extracted from the preprocessed signal, frequency domain feature fusion is performed, and a fusion feature vector is generated. Specifically, the module uses fast Fourier transform to perform frequency domain conversion on the preprocessed signal, and respectively extracts the human feature spectrum and the obstacle echo intensity spectrum. The human feature spectrum matches the feature template through a pre-trained model, identifies typical human motion frequency and micro-motion features, and enhances the recognition degree of human signals; the obstacle echo intensity extracts the reflection intensity information through the energy integration and peak detection algorithm, and distinguishes static and dynamic obstacles. Subsequently, combined with the multi-modal data fusion algorithm, a weighted fusion strategy is used to fuse the human feature spectrum and the obstacle echo intensity, and the weight coefficient is dynamically adjusted according to the environmental interference level. During the fusion process, the system introduces an interference elimination module, which eliminates false echoes through adaptive filtering and outlier rejection for multipath effect and environmental noise. The fusion feature vector is normalized to unify the dimension, which is convenient for subsequent analysis. The fusion feature vector is taken as the output field and is transmitted to the “to-be-analyzed data” of S130, providing basic data for subsequent adaptive filtering processing.

[0105] S130, adaptive Kalman filtering is performed on the fusion feature vector to generate standardized fusion data.

[0106] The adaptive Kalman filtering is performed on the fusion feature vector to generate standardized fusion data. Specifically, the module constructs a state space model according to the fusion feature vector, and uses an extended Kalman filter algorithm to dynamically estimate and correct errors for multi-modal data. First, the system initializes the filter parameters, including the state transition matrix, the observation matrix and the noise covariance matrix, and performs parameter adaptive adjustment combined with historical data. Subsequently, the prediction step is executed, the current state is predicted based on the state estimation of the last time, and the fusion feature vector observed is updated to correct the state estimation value. During the filtering process, the system dynamically adjusts the noise covariance matrix to optimize the filtering performance for environmental changes and radar signal uncertainty. The filtering result is standardized to unify the data scale and format, and the standardized fusion data is formed. The system detects and compensates for abnormal states and data loss during the filtering process to ensure data continuity and accuracy. Finally, the standardized fusion data is taken as the output field and is transmitted to the “input data” of S210 for use by the target type and distance identification module, and the high-precision fusion processing of multi-modal radar signals is completed.

[0107] Step S200 includes at least steps S210-S230:

[0108] S210, obtaining fusion data, performing convolutional neural network target detection, and obtaining preliminary classification results.

[0109] The fusion data is obtained as input data, specifically, the fusion data is output by a multi-modal radar signal fusion processing module S130, and contains multi-channel radar information subjected to adaptive Kalman filtering standardization processing. The module first inputs the fusion data into a convolutional neural network target detection sub-module through a high-speed bus interface. The convolutional neural network is based on a pre-trained model combined with a multi-layer feature extraction structure, and performs multi-scale spatial feature analysis and time sequence feature capture on the input fusion data. Specifically, the convolutional neural network adopts a multi-layer convolutional layer and a pooling layer stacked alternately, combined with batch normalization and an activation function processing, to effectively extract human motion features and obstacle contour features. Further, the network outputs a multi-class probability distribution through a Softmax layer, to realize preliminary classification of human bodies, static obstacles and dynamic obstacles and the like. To ensure the real-time performance and accuracy of detection, the module sets a dynamic threshold filtering mechanism to eliminate low-confidence targets and prevent false positives. The preliminary classification result contains a target class label and a corresponding confidence, which is recorded in a target identification log by the system for subsequent performance evaluation and model updating. Finally, the target classification result is transmitted as an output field to the "to-be-verified data" of S220, for use by a target distance accurate calculation module, to realize a closed-loop data flow from fusion data to target classification.

[0110] S220, extracts distance parameters from the preliminary classification result, performs Doppler shift compensation, and generates an accurate distance value.

[0111] The distance parameters are extracted from the target classification result, specifically, the module first analyzes spatial positioning information and speed estimation data in the target classification result as basic input of the distance parameters. Then, a Doppler shift compensation algorithm is combined to correct the frequency offset in the target echo signal to eliminate the influence of frequency drift caused by the motion of the sightseeing vehicle itself or the target dynamics. The Doppler shift compensation algorithm uses a frequency spectrum analysis method based on fast Fourier transform, combined with real-time speed sensor data, to dynamically adjust the compensation coefficient. Further, the system calculates the accurate distance value between the target and the sightseeing vehicle through time delay ranging technology combined with the time difference measurement of the multi-modal radar. To improve the accuracy of distance calculation, the module performs multiple sampling and weighted average processing on the ranging result, eliminates outliers, and adjusts the filtering parameters according to the environmental noise level. The system monitors the stability of the distance data in real time during processing, and triggers the resampling mechanism when there is an abnormal fluctuation, to ensure the reliability of the ranging result. Finally, the target distance value is transmitted as an output field to the "distance parameter" of S230, to provide an accurate distance basis for target identification result generation.

[0112] S230, type-distance association matching is performed on the distance value to generate a target identification result.

[0113] The target distance value is obtained as an input distance parameter. Specifically, the module first matches the target distance value with the target classification result for type-distance association, combines a preset target type distance threshold database, and realizes accurate identification of the target. The type-distance association matching is performed by constructing a multi-dimensional association matrix to map the target category label and the corresponding distance range to identify the spatial distribution characteristics of the human target and the obstacle. Further, the system uses a rule engine to logically determine the matching result to distinguish between human and non-human obstacles, and combines target trajectory analysis to determine the dynamic or static properties of the target. The rule engine is based on an expert experience knowledge base, combined with real-time environmental parameters such as sightseeing vehicle speed, radar detection angle, and environmental interference coefficient, to dynamically adjust the matching strategy. To ensure the timeliness of the identification result, the module sets a time window mechanism to continuously track the target state change and update the identification result. The system records abnormal matching conditions in the identification process in an exception log for subsequent diagnosis. Finally, the target identification result is output as an input parameter to the "input parameter" of S310 to provide accurate target information support for the hierarchical response strategy generation module.

[0114] Step S300 includes at least steps S310-S330:

[0115] S310, obtain the target identification result, and perform 8m / 1.5m double threshold comparison to generate a preliminary response level.

[0116] The target identification result is obtained as an input parameter. Specifically, the target identification result is output from the previous step S230 and transmitted to the present step through a high-speed data bus. The system first performs data format verification and integrity detection on the target identification result to ensure that the input data is complete and normal. Then, based on the target type information and distance information in the target identification result, the system performs double threshold comparison processing, specifically comparing the target distance value with the preset 8m and 1.5m thresholds one by one. The 8m threshold corresponds to the long-distance early warning range of the microwave radar human sensor, and the 1.5m threshold corresponds to the short-distance obstacle detection range of the ultrasonic radar. The system sets logical judgment rules, and when the target distance is less than or equal to 8m and the target type is human, it is determined as a first-level response condition; when the target distance is less than or equal to 1.5m and the target type is an obstacle, it is determined as a second-level response condition. The comparison process uses a hardware timing trigger mechanism to ensure real-time and response speed, and combines software algorithms to filter multiple consecutive frames of data to reduce the probability of false positives. The system assigns a preliminary response level to the comparison result, specifically, the first-level response corresponds to a speed limit warning level, the second-level response corresponds to an emergency braking level, and if neither is met, the response level is no response. The response level is transmitted as an output field to the "to-be-optimized parameter" of the next processing step S320 for subsequent response instruction optimization processing.

[0117] S320, extracting the environmental interference coefficient from the response level, performing dynamic weight adjustment, and generating an optimized response instruction.

[0118] Further, the environmental interference coefficient is extracted from the response level as the basis for dynamic weight adjustment. Specifically, the system obtains the current environmental light intensity, temperature change, and radar signal noise level through the environmental monitoring module, combines historical data and real-time sampling, and calculates the environmental interference coefficient. The environmental interference coefficient reflects the influence of the current environment on the radar signal, with a numerical range of 0 to 1. The larger the value, the stronger the interference. The system inputs the response level and the environmental interference coefficient into the dynamic weight adjustment algorithm, which is designed based on a fuzzy logic controller. The algorithm combines the urgency of the response level and the change trend of the environmental interference to dynamically adjust the weight parameters of the response instruction. Specifically, the algorithm assigns different confidence levels to the response level through a weight function, enhancing the response accuracy in high-interference environments while avoiding false triggers caused by environmental abnormalities. During the adjustment process, the system uses a sliding window mechanism to smooth the weight changes, preventing instruction fluctuations caused by sudden environmental changes. Finally, the system generates an optimized response instruction based on the adjusted weight parameters, including the response level, priority identifier, and execution time window. The optimized instruction is output as the "to-be-confirmed instruction" in the next processing step S330, which is used for subsequent timing conflict detection and strategy generation.

[0119] S330, timing conflict detection on the optimized response instruction, and generation of a final hierarchical response strategy.

[0120] The timing conflict detection of the optimization instruction is a key link for generating the final hierarchical response strategy. Specifically, the system receives the optimization instruction from S320, and combines the current system state and historical response records to construct a timing event sequence. The timing conflict detection module uses a state machine-based algorithm model to identify time overlap, priority conflict, and execution resource competition between instructions. The system first sorts the time window of the optimization instruction, detects whether there is an overlapping interval of instruction execution time, and if overlap is found, further analyzes the instruction priority. The instruction with high priority obtains the execution right, and the low-priority instruction is delayed or canceled. During the detection process, the system combines the relay execution state feedback to confirm the current hardware execution capability, avoiding hardware damage or response failure caused by instruction conflict. For the detected abnormal conflict situation, the system automatically triggers the conflict resolution mechanism, including instruction rescheduling, priority re-arbitration, and temporary response level adjustment. The conflict resolution strategy is based on a preset rule base and an online learning algorithm, which can adapt to different environments and operating states. Finally, the system generates a complete and conflict-free hierarchical response strategy, which includes response level, execution time, priority, and control parameters, ensuring the accuracy of the response instruction and the reliability of the execution. The hierarchical strategy is used as an output field and transmitted to the "control parameters" of the subsequent step S410 for the hardware control signal generation module, completing the closed-loop control from target identification to hardware response.

[0121] Step S400 includes at least steps S410-S430:

[0122] S410, obtain a hierarchical response strategy, convert PWM signals, and generate a primary control waveform.

[0123] The hierarchical strategy is obtained, specifically, the hierarchical strategy is output by the hierarchical response strategy generation module of S330 as the input data of the present step S410, and the hierarchical strategy contains control parameters and instruction priority information for different response levels. First, the system accesses the hierarchical strategy data through a digital signal processing interface, performs format analysis and field verification, and ensures the integrity and consistency of the hierarchical strategy. For each level of response instruction in the hierarchical strategy, the system maps the discrete response level to the corresponding PWM (Pulse Width Modulation, PWM) duty cycle and frequency parameters according to the preset PWM signal conversion rule, specifically, a low-level response corresponds to a PWM signal with a small duty cycle, and a high-level response corresponds to a PWM signal with a large duty cycle, to realize fine adjustment of the subsequent hardware control signal. The PWM signal conversion process cooperates with the hardware timer and the digital signal generation module, the timer accurately controls the PWM period according to the system clock period, and the digital signal generation module adjusts the duty cycle according to the mapping rule to ensure the stability and accuracy of the PWM signal. During the signal generation process, the system monitors the amplitude and frequency of the PWM waveform in real time, and uses a closed-loop feedback mechanism to fine-tune the waveform to avoid waveform distortion caused by hardware jitter or electromagnetic interference. Further, the module sets an abnormality detection mechanism, when the PWM waveform exceeds the preset amplitude or frequency range, an error flag is triggered and an abnormal event log is recorded for subsequent diagnosis and maintenance. After the PWM signal conversion is completed, the module generates the primary control waveform as the output field and transmits it to the "to-be-verified signal" of S420 for further processing by the relay drive adaptation module, and at the same time, the control waveform data is collected by the system state monitoring unit for subsequent state feedback control reference.

[0124] S420, extract the voltage amplitude from the primary control waveform, perform relay drive adaptation, and generate a drive signal.

[0125] Further, the voltage amplitude is extracted from the control waveform, specifically, the S420 module receives the primary control waveform from S410 as input, converts the PWM waveform to voltage signal samples through an analog signal acquisition interface, and completes digitization sampling using a high-precision analog-to-digital converter (ADC). The sampling process uses synchronous sampling technology to ensure that the sampling time is strictly aligned with the PWM signal period, avoiding sampling deviation. The voltage amplitude data obtained by sampling is first processed by a digital filter, using a finite impulse response (FIR) filter to eliminate high-frequency noise and transient interference and improve signal quality. Subsequently, the system converts the processed voltage amplitude into the level parameter of the relay drive signal through a voltage amplitude mapping algorithm, specifically, according to the relay drive voltage threshold, the high and low level states of the drive signal are determined to meet the starting and maintaining requirements of the relay. The mapping algorithm considers factors such as environmental temperature and power fluctuations, dynamically adjusts the voltage threshold, and ensures the reliability and stability of the relay drive. To adapt to different models of relays, the system has multiple built-in drive protocol parameters, and the relay drive adaptation module automatically selects the corresponding protocol to generate the drive signal according to the control parameters in the hierarchical strategy, realizing compatible control of the motor controller. During the generation of the relay drive signal, the system introduces a redundancy checking mechanism, using dual-channel signal generation and comparison to discover and correct drive signal abnormalities in a timely manner, preventing misoperation. After the drive signal is generated, the module transmits the drive signal as an output field to the "to-be-verified instruction" of S430 for subsequent hardware compatibility detection, while the drive signal is collected by the real-time monitoring unit and fed back to the system state feedback control module.

[0126] S430, hardware compatibility detection is performed on the drive signal to generate a control instruction.

[0127] The driving signal is acquired, S430, which is a terminal processing unit for generating a hardware control signal. Specifically, the driving signal from S420 is received as input, and hardware compatibility detection is first performed. The detection process includes signal level matching, timing compliance checking, and electrical parameter verification. The detection is completed by a built-in hardware interface test unit. The test unit uses an oscilloscope sampling module to sample the driving signal in real time, collects the voltage, current, and timing parameters of the signal, and compares them with the technical specifications of the relay and motor controller one by one. For signal level matching, the system ensures that the high and low levels of the driving signal meet the minimum and maximum thresholds of the relay driving voltage, respectively. Timing compliance checking verifies whether the period, duty cycle, and switching speed of the PWM signal meet the hardware response requirements. Electrical parameter verification covers load capacity, anti-interference ability, and signal stability. During the detection process, the system uses multi-point sampling and statistical analysis methods to identify occasional abnormalities and persistent faults, and realizes comprehensive evaluation of hardware compatibility. If the detection result finds abnormalities, the system automatically calls the preset fault handling program to adjust the driving signal parameters or switch to the backup signal channel, ensuring effective transmission of control commands. Further, the module combines redundancy check information to verify the integrity of the driving signal, preventing data loss or error codes during signal transmission. After completing compatibility detection and verification, the system generates and encapsulates the final control command as an output field, which is transmitted to "execution parameters" of S510 for use by the relay execution control module. At the same time, the status of the final control command and the detection log are recorded in the system log management unit for subsequent maintenance and performance analysis. The control signal link formed by S410 to S430 realizes seamless conversion and verification from hierarchical strategy to hardware execution signal, ensuring the accuracy and reliability of relay driving.

[0128] Step 500 includes at least steps S510-S530:

[0129] S510, acquire the final control command, switch the relay contact state, and generate the primary execution signal.

[0130] The control instruction is obtained, specifically, the control instruction is generated by the foregoing S430 step, and is accessed into the processing flow of the relay execution control module S500 as input. The control instruction is first received by the system through a multi-input interface, and is subjected to format analysis and parameter mapping according to a preset relay driving protocol. The system strictly checks the voltage level, pulse width modulation (PWM) signal characteristics and time sequence in the control instruction, to ensure that the control instruction meets the electrical specifications and time sequence requirements of the relay hardware. Specifically, a digital signal processor is used to sample the input signal in real time, and a hardware interrupt mechanism is combined to realize fast response of the control instruction. Subsequently, the system determines whether the execution condition is met according to the contact state of the relay and the current execution environment, to avoid misoperation of the relay due to repeated instructions or conflicting instructions. The determination is realized through a state machine model, and a redundancy detection mechanism is combined to ensure the uniqueness and accuracy of the instruction. In the execution process, the system performs closed-loop monitoring on the switching action of the relay contact, uses a built-in current sensor and a voltage detection module to collect contact state information in real time, and determines whether the relay is successfully switched. In an abnormal case, the system automatically triggers a fault alarm, and records abnormal information in a system log for subsequent maintenance and diagnosis. Finally, the step outputs an execution signal as input of “to-be-responded data” of the next step S520, to ensure the signal integrity and accuracy of the relay execution control, and to provide basic data for subsequent execution effect verification.

[0131] S520, extracting motor response data from the primary execution signal, performing execution effect verification, and generating an execution result.

[0132] Extract motor response data from the execution signal, specifically, the execution signal is accessed as input in S520 step, starting to collect and analyze the motor response triggered by the relay action. The system acquires real-time motor state information through the motor controller communication interface, including but not limited to speed feedback, current consumption, torque change and response delay, etc. Key parameters. The motor response data is collected by high-speed analog-to-digital converter, combined with time synchronization mechanism to ensure the timing accuracy of the data. Further, the system uses filtering algorithm to preprocess the collected motor response signal, eliminates environmental noise and electromagnetic interference, and improves the effectiveness and stability of the data. Then, according to the preset execution effect evaluation model, combined with historical execution data, the system comprehensively evaluates the motor response through multi-dimensional indicators to judge whether the execution effect of the relay control instruction meets the expectation. The evaluation process includes response time detection, action amplitude comparison and abnormal waveform identification, ensuring the accuracy and safety of motor control. For the detected abnormalities or deviations, the system automatically generates an execution exception report and notifies the upper control module through the internal communication bus for adjustment. The execution result is passed to the "feedback parameter" input of S530 as an output field, supporting the timing marking and closed-loop control of subsequent execution records, forming a closed-loop verification mechanism for execution effect.

[0133] S530, time sequence marking of execution result, generating execution record with timestamp.

[0134] Time sequence marking of the execution result, specifically, the execution result is called as input in S530 step, the system first synchronizes the timestamp of the execution result data stream, combined with the global clock source to ensure the timing consistency of all execution events. The timestamp is generated by a high-precision real-time clock module, and the clock is calibrated through a synchronization protocol with the control parameters of the multi-modal radar signal fusion processing module S100, ensuring the uniformity of time reference among system modules. Then, the system marks the key event nodes in the execution result, including the starting time of relay contact switching, the completion time of execution and the occurrence time of abnormal events. The marking process is automatically triggered through event-driven mechanism, combined with buffer management strategy to ensure the integrity and continuity of timing information. Further, the system structures the execution record, including execution signal parameters, execution effect evaluation and corresponding timestamp, forming a traceable execution history. This execution record is not only used for diagnosis and maintenance within the module, but also transmitted to the system state feedback control module S600 through the data bus as input data for brake release condition detection and speed recovery control. Finally, the execution record is passed to the "input data" input of S610 as an output field, completing the closed-loop information feedback of the relay execution control link, ensuring the real-time and reliability of the sightseeing vehicle anti-collision control system.

[0135] Step S600 includes at least steps S610-S630:

[0136] S610, obtain the execution record, perform brake release condition detection, and generate a state determination result.

[0137] The execution record is obtained as an input source. Specifically, the execution record is output by step S530 of the relay execution control module S500 and includes time-stamped execution signal states and execution effect data. The system first accesses the execution record through a communication interface, performs format analysis and integrity verification, and ensures data integrity and compliance with the preset protocol specification. Subsequently, the system analyzes the relay contact state changes, motor response indicators, and time sequence information contained in the execution record according to a predefined brake release condition rule library. Specifically, the system determines whether the current execution state meets the conditions for releasing the brake, such as the relay contact returning to normal, the execution signal being stable and maintaining a certain time window without abnormal fluctuations, etc. Further, the system combines environmental perception parameters and historical execution records, uses a multi-threshold determination mechanism, and dynamically adjusts the brake release trigger threshold to prevent false positives due to short-term abnormal signals. For abnormal or missing data, the system starts a redundancy verification process, compensates and eliminates the data through data interpolation and anomaly detection algorithms, and ensures the reliability of the determination data. The state determination result is logically synthesized to form structured state determination information, which specifically includes brake release state flags, abnormal warning identifiers, and time stamps. The system transmits the state determination result as an output field to the "adjustment parameters" of S620 for subsequent speed recovery and control adjustment by the system state feedback control module.

[0138] S620, extract the speed recovery parameters from the state determination result, perform PID control calculation, and generate an adjustment instruction.

[0139] The speed recovery parameters are extracted from the state determination, PID control calculation is performed, and an adjustment instruction is generated. Specifically, the system first determines whether the current speed recovery stage is entered according to the brake release state flag contained in the state determination. Then, the system obtains speed recovery parameters related to the current vehicle speed, target speed and execution error from the state determination result, including the current vehicle speed, target speed limit value, execution signal response delay and environmental disturbance coefficient. The PID controller module performs proportional, integral and differential calculations based on the above parameters to calculate an adjustment instruction that adapts to the current operating environment. Specifically, the proportional term directly adjusts the control strength according to the speed error, the integral term accumulates historical errors to eliminate steady-state deviation, and the differential term predicts future trends to suppress overshoot and oscillation. Further, the system dynamically adjusts the PID parameters in combination with the environmental disturbance coefficient to enhance the robustness and response speed of the control. During the control calculation process, the system monitors the output range of the control instruction in real time to ensure that the adjustment instruction is within the safety boundary, preventing over-adjustment or under-adjustment from causing safety hazards. For abnormal input or sudden events, the system starts the fault protection mechanism and automatically switches to the preset safety mode to ensure stable operation of the vehicle. Finally, the adjustment instruction is output as a field to the "closed-loop parameters" of S630, providing input for system parameter synchronization and feedback configuration to complete the closed-loop control calculation of speed recovery.

[0140] S630, synchronizing system parameters to the adjustment instruction to generate a closed-loop feedback configuration.

[0141] Synchronization of system parameters to the adjustment instruction to generate a closed-loop feedback configuration. Specifically, the system first receives the adjustment instruction from S620 as the core input of the closed-loop control, and then maps the instruction to multiple control parameters within the system, including but not limited to actuator drive voltage, relay trigger timing and alarm signal trigger conditions. The system executes parameter synchronization strategies through the parameter management module based on the current vehicle operating state and historical feedback data to ensure the consistency and real-time performance of the adjustment instruction among the system sub-modules. Specifically, a timestamp synchronization mechanism is used to coordinate the action timing of each control unit to prevent control failure caused by instruction delay or misplacement. Further, the system executes a feedback configuration generation process to dynamically adjust the feedback loop gain and filter parameters based on historical data from the execution record to optimize the closed-loop response performance. For abnormal or conflicting parameters in the feedback configuration, the system automatically performs conflict detection and priority arbitration to ensure the stability and safety of the closed-loop control. The closed-loop feedback configuration is broadcast to the S110 step of the multi-modal radar signal fusion processing module S100 through the control bus as the "control parameter" input, achieving dynamic adjustment of multi-modal radar signal acquisition and preprocessing to form a complete closed-loop control system. Finally, the feedback configuration is output as a field to complete the closed-loop feedback configuration generation of the system state feedback control module, ensuring continuous, stable and efficient operation of the sightseeing vehicle collision avoidance control system.

[0142] The steps S610 to S630 realize the closed-loop linkage between the relay execution control and the multi-modal radar signal fusion processing by the brake release condition detection of the execution record, the speed recovery adjustment based on the PID control, and the synchronous feedback configuration of the system parameters, effectively guaranteeing the dynamic response and safety and stability of the sightseeing vehicle anti-collision system.

[0143] In another embodiment:

[0144] S610, acquire the execution record, detect the brake release condition, and generate a state determination result.

[0145] The execution record is acquired as an input source. Specifically, the execution record is output by the step S530 of the relay execution control module S500 and contains the execution signal state and execution effect data with a time stamp. After the system accesses the execution record through a communication interface, performs format analysis and integrity verification, and based on a preset brake release condition rule library, the relay contact state change, motor response index, and time sequence information are subjected to multi-threshold determination.

[0146] Formula ① defines the dynamic adjustment mechanism of the brake release trigger threshold value:

[0147]

[0148] Among them:

[0149] : dynamic brake release threshold value;

[0150] , , : weight coefficient;

[0151] : basic threshold value;

[0152] : energy gradient of the kth execution signal;

[0153] : the kth execution signal; : environmental disturbance factor;

[0154] : total number of environmental disturbance factor categories;

[0155] : environmental disturbance intensity index.

[0156] Explanation: Formula ① dynamically adjusts the brake release threshold value according to the voltage change gradient of the execution record and the environmental disturbance parameters , ensuring that the threshold value adapts to the current environment and execution state.

[0157] Further, formula ② is used for logical synthesis of state determination:

[0158]

[0159] Wherein:

[0160] : Brake release state flag, calculated by formula ① and the actual contact voltage in the 'execution record' ;

[0161] : Activation function;

[0162] : Normalization coefficient, dynamically adjusted according to historical data standard deviation;

[0163] : Multiplication operator; is the index variable;

[0164] : The confidence weight of the first historical execution record, derived from the number of successful executions in the 'execution record';

[0165] : Historical execution matching degree, obtained by time series similarity analysis;

[0166] : Historical execution record number, upper limit of index.

[0167] Note: Formula ② takes dynamic threshold and actual voltage as core inputs, combines historical confidence and matching degree, and outputs brake release state , providing key basis for subsequent control.

[0168] S610→S620: S610 module dynamically calculates brake release threshold according to execution record and environmental parameters , and outputs state determination result . As input, it is passed to S620 to determine the speed error and the starting condition of PID control.

[0169] S620, extract speed recovery parameters from state determination results, perform PID control calculation, and generate adjustment instructions.

[0170] S620, extract speed recovery parameters from state determination results, perform PID control calculation, and generate adjustment instructions. S620, extract speed recovery parameters from state determination results, perform PID control calculation, and generate adjustment instructions.

[0171] Equation ③ builds the proportional-integral-derivative term:

[0172]

[0173] wherein:

[0174] PID control output instruction, used to adjust the motor driving signal;

[0175] Speed error, calculated by the target speed limit value in the "state determination result" and the current vehicle speed

[0176] , , Dynamic PID coefficient, adjusted according to the environmental disturbance coefficient , , Environmental disturbance coefficient;

[0177] Integral time interval.

[0178] Further, Equation ④ defines the safety boundary constraint of the control instruction:

[0179]

[0180] wherein:

[0181] Final adjustment instruction;

[0182] PID output change gradient;

[0183] Gradient change threshold;

[0184] , Safety range of control instruction, set according to the motor controller specification;

[0185] Amplitude limiting function;

[0186] L2 norm.

[0187] The adjustment instruction output by Equation ④ is input to the S630 module to generate a closed-loop feedback configuration, achieving time-domain synchronization and optimization of control parameters.

[0188] ​S620→S630: S620 module performs PID control based on state determination results, calculates control output And through the security boundary constraint, the final adjustment instruction is generated . The adjustment instruction is passed as input to S630 for time domain parameter synchronization and closed loop feedback gain optimization.

[0189] S630, system parameter synchronization is performed on the adjustment instruction to generate a closed loop feedback configuration.

[0190] S620 is connected to , to generate a closed loop feedback configuration. Formula ⑤ defines the time domain mapping of parameter synchronization:

[0191]

[0192] Wherein:

[0193] : the synchronized actuator driving voltage, calculated by weighting and historical execution timestamp ;

[0194] : time decay weight, calculated according to , wherein is the current and historical time difference;

[0195] : decay coefficient, dynamically adjusted according to the acceleration in the'vehicle state data';

[0196] : current time;

[0197] : the th historical execution timestamp;

[0198] : the number of historical execution records.

[0199] Further, formula ⑥ realizes feedback loop gain optimization:

[0200]

[0201] Wherein:

[0202] : optimized loop gain;

[0203] : regularization coefficient, derived from the 'threshold and strategy table';

[0204] : gradient of synchronized voltage to gain;

[0205] : state determination gradient;

[0206] : regularization coefficient;

[0207] : L2 norm;

[0208] : L1 norm;

[0209] : minimum value of a specific G value;

[0210] : loop gain original variable to be optimized.

[0211] The output field of this paragraph is a closed-loop feedback configuration, which includes and is transmitted to the "control parameter" input of S100 through the control bus. S630→S100, the S630 module calculates the synchronous actuator driving voltage by historical time weighting and dynamic attenuation, and optimizes the loop gain in combination with the state determination gradient of S610. The generated closed-loop feedback configuration is transmitted to the S100 module through the control bus, completing the backfilling of the control parameters and realizing system closed-loop control.

[0212] Technical effect: This paragraph realizes dynamic feedback adjustment of control instructions to multi-modal radar signal acquisition through time domain parameter synchronization and loop gain optimization, forms a closed-loop control link, and ensures the consistency and environmental adaptability of system parameters. S610-S630 constructs a complete feedback link from execution record analysis to radar signal adjustment through threshold dynamic adjustment, PID control optimization and closed-loop parameter synchronization, improving the response accuracy and stability of the system.

[0213] Embodiment two: Figure 2 A structural block diagram of a sightseeing vehicle anti-collision control system based on multi-modal radar according to an embodiment of the present application is shown. As Figure 2 shown, the structure can include:

[0214] The installation and calibration unit 01 is used to obtain multi-modal radar installation parameters and complete sensor calibration. Specifically, installation position information and sensor initial parameters from the sightseeing vehicle system are received, and under the configured installation constraints, the physical installation of the multi-modal radar and the calibration process of the sensor are completed, forming the installation parameters and calibration data of the multi-modal radar; the installation parameters and calibration data are recorded as calibration records and kept consistent with the association of the sensor state; the calibration records are transmitted to the sector limiting unit as input parameters, while the calibration log is reserved for subsequent tracing and updating.

[0215] Sector defining unit 02, for setting the horizontal field of view angle coverage range of the millimeter wave radar. Specifically, receiving the calibration record and the sightseeing vehicle driving environment data from the installation and calibration unit output, performing field of view angle calculation, sector division and coverage range calibration on the calibration record, forming the field of view angle sector defining information of the millimeter wave radar according to the configuration parameters; delivering the field of view angle sector defining information as sector data to the threshold and strategy configuration unit, and registering the corresponding environment state information in the storage for the threshold and strategy configuration unit to read.

[0216] Threshold and strategy configuration unit 03, for configuring the collision risk judgment threshold and the dynamic arbitration strategy table. Specifically, receiving the sector data from the sector defining unit, performing threshold calculation and strategy table configuration in combination with the driving speed and path information of the sightseeing vehicle, generating the collision risk judgment threshold and the dynamic arbitration strategy table when the safety standard condition is met; delivering the collision risk judgment threshold and the dynamic arbitration strategy table to the data acquisition unit as strategy parameters, and recording the corresponding relationship with the environment parameters in the configuration storage.

[0217] Data acquisition unit 04, for acquiring the ranging and speed measurement data of the millimeter wave radar and the point cloud data of the laser radar. Specifically, completing data acquisition configuration based on the strategy parameters from the threshold and strategy configuration unit, generating data acquisition instructions; outputting the data acquisition instructions to the millimeter wave radar and laser radar interface, collecting ranging and speed measurement data and point cloud data, and returning the data acquisition state to the time synchronization unit for registration.

[0218] Time synchronization unit 05, for aligning the timestamps of the multi-modal radar and performing spatial coordinate mapping. Specifically, receiving the data acquisition state and the ranging and speed measurement data, point cloud data from the data acquisition unit, performing timestamp alignment and coordinate system mapping to obtain fused multi-modal data; providing the fused multi-modal data to the event judgment unit as input data, and maintaining an index relationship consistent with the time sequence.

[0219] Event judgment unit 06, for generating a collision risk level based on the fused target trajectory sequence. Specifically, receiving the fused multi-modal data from the time synchronization unit, analyzing and risk assessing the target trajectory, generating a collision risk level according to the judgment rules; delivering the collision risk level as risk data to the arbitration unit, and recording the corresponding relationship between the risk level and the event record in the risk level.

[0220] Arbitration unit 07, for generating hierarchical control instructions according to the priority strategy table decision. Specifically, completing the lookup and decision of the priority strategy table based on the risk data from the event judgment unit, generating hierarchical control instructions; outputting the hierarchical control instructions to the speed limit signal output unit and the brake signal output unit, and returning the decision state to the log and parameter update unit for registration.

[0221] The speed limiting signal output unit 08 is configured to send a speed reducing instruction to the power controller. Specifically, the hierarchical control instruction from the arbitration unit is received, the speed reducing control information is parsed to generate the speed reducing instruction, the speed reducing instruction is output to the power controller, and the effective state is fed back to the log and parameter updating unit for registration.

[0222] The brake signal output unit 09 is configured to send an emergency brake instruction to the brake actuator. Specifically, the hierarchical control instruction from the arbitration unit is received, the brake control information is parsed to generate the emergency brake instruction, the emergency brake instruction is output to the brake actuator, and the effective state is fed back to the log and parameter updating unit for registration.

[0223] The alarm management unit 10 is configured to trigger the audible and light alarm device. Specifically, the hierarchical control instruction from the arbitration unit is received, the alarm information is parsed to generate the alarm trigger signal, the alarm trigger signal is output to the audible and light alarm device, and the trigger state is fed back to the log and parameter updating unit for registration.

[0224] The log and parameter updating unit 11 is configured to record the control instruction timestamp and update the strategy table parameters. Specifically, the effective states from the speed limiting signal output unit, the brake signal output unit and the alarm management unit, and the strategy parameters from the threshold and strategy configuration unit are received, log recording, timestamp registration and strategy table parameter updating are performed to obtain the updated log record and strategy table, the updated log record and strategy table are provided to the installation and calibration unit as parameter backfill, and the index relationship consistent with the version is maintained.

Claims

1. A sightseeing vehicle anti-collision control method based on multi-modal radar, characterized in that: include: 24G microwave radar and ultrasonic radar raw signals are collected, timestamp alignment and noise reduction are performed, and standardized fusion data is generated. The raw signals are connected to the system through a multi-channel data acquisition interface. The system uses a preset sampling frequency and time synchronization mechanism. A combination of hardware triggering and software timing is used to capture the signal sampling moment in real time, and a timestamp matching algorithm is used to eliminate abnormal time series data. Combining adaptive filtering and wavelet transform technology, it suppresses high-frequency noise and low-frequency drift components respectively; identifies noise peaks through a dynamic threshold adjustment mechanism, and performs interpolation correction on abnormal signal segments; Based on the fused data, multi-target classification is achieved through convolutional neural networks, Doppler shift compensation and distance calculation are performed, and target recognition results are output; Perform dual-threshold comparison and dynamic weight adjustment of environmental interference on target recognition results to complete timing conflict detection and generate hierarchical response strategies; Convert it into PWM control signal according to the hierarchical response strategy, extract the voltage amplitude to complete relay drive signal adaptation and hardware compatibility detection, and generate control instructions; Receive control instructions, drive relays and collect motor responses in real time, complete execution effect evaluation and timing marking, and generate execution records with timestamps; Analyze execution records to determine brake release conditions, generate adjustment instructions based on PID control, and feedback to configure closed-loop parameters to achieve system parameter synchronization and closed-loop control; The expressions of the feedback configuration closed-loop parameters include: in, is the optimized loop gain; is the regularization coefficient; is the gradient of the synchronization voltage to the gain; is the state determination gradient; is the regularization coefficient; is the L2 norm; is the L1 norm; is the specific G value that takes the minimum value; is the original variable of the loop gain to be optimized.

2. The sightseeing vehicle collision avoidance control method based on multimodal radar according to claim 1, characterized in that: Based on the fused data, a convolutional neural network is used to achieve multi-target classification, perform Doppler shift compensation and distance calculation, and output target recognition results, including: The fused data is connected to the convolutional neural network target detection submodule through a high-speed bus interface. The convolutional neural network is based on a pre-trained model combined with a multi-layer feature extraction structure to perform multi-scale spatial feature analysis and time series feature capture on the input fused data. The convolutional neural network uses multiple layers of convolutional layers and pooling layers stacked alternately, combined with batch normalization and activation function processing, to extract human motion features and obstacle contour features; The Softmax layer outputs multi-category probability distribution to achieve preliminary classification of human bodies, static obstacles, and dynamic obstacles. Set up a dynamic threshold filtering mechanism to eliminate low-confidence targets and prevent false alarms; The target classification results are recorded in the target recognition log for subsequent performance evaluation and model update.

3. The sightseeing vehicle collision avoidance control method based on multimodal radar according to claim 1, characterized in that: Based on the fused data, multi-target classification is achieved through convolutional neural networks, Doppler shift compensation and distance calculation are performed, and the process of outputting target recognition results also includes: Parse the spatial positioning information and speed estimation data from the target classification results as the basic input for the distance parameter; Combined with the Doppler frequency shift compensation algorithm, the frequency offset in the target echo signal is corrected to eliminate the frequency drift caused by the sightseeing vehicle's own movement or the target's dynamics; The Doppler shift compensation algorithm uses a spectrum analysis method based on fast Fourier transform, combined with real-time speed sensor data, to dynamically adjust the compensation coefficient; The precise distance between the target and the sightseeing vehicle is calculated by using time delay ranging technology combined with the time difference measurement of the multi-modal radar. The ranging results are sampled multiple times and weighted averaged to eliminate outliers, and the filtering parameters are adjusted according to the environmental noise level.

4. The sightseeing vehicle collision avoidance control method based on multimodal radar according to claim 1, characterized in that: Perform dual-threshold comparison and dynamic weight adjustment of environmental interference on the target recognition results, specifically comparing the target distance value with the preset 8-meter and 1.5-meter thresholds one by one; When the target distance is less than or equal to 8 meters and the target type is human, it is determined to be a level 1 response condition; When the target distance is less than or equal to 1.5 meters and the target type is an obstacle, it is determined to be a secondary response condition; Adopt hardware timing trigger mechanism and combine software algorithm to filter continuous multi-frame data to reduce the probability of misjudgment; The comparison results are used to assign a preliminary response level. The first-level response corresponds to the speed limit warning level, and the second-level response corresponds to the emergency braking level. If neither is met, the response level is no response.

5. The sightseeing vehicle collision avoidance control method based on multi-modal radar according to claim 1, characterized in that: The process of dynamic weight adjustment of environmental interference includes: The environmental monitoring module obtains the current environment's light intensity, temperature changes, and radar signal noise level, and calculates the environmental interference coefficient by combining historical data and real-time sampling; The response level and environmental interference coefficient are input into the dynamic weight adjustment algorithm, and the weight parameters of the response instructions are dynamically adjusted based on the fuzzy logic controller design; A sliding window mechanism is used to smooth weight changes to prevent sudden environmental changes from causing instruction fluctuations; An optimized response instruction is generated based on the adjusted weight parameters, and the instruction content includes a response level, a priority identifier, and an execution time window.

6. The sightseeing vehicle anti-collision control method based on multi-modal radar according to claim 1, characterized in that: The process of completing timing conflict detection and generating a hierarchical response strategy specifically includes: Construct a sequence of timing events to identify time overlaps, priority conflicts, and execution resource contention between instructions; Sort the time windows of the optimized instructions and detect whether there are overlapping intervals of instruction execution time; If overlap is found, the instruction priority is further analyzed, and the instruction with higher priority is executed, while the instruction with lower priority is delayed or cancelled; Combined with relay execution status feedback, confirm the current hardware execution capability to avoid hardware damage or response failure caused by instruction conflicts; For abnormal conflict situations detected, the conflict resolution mechanism is automatically triggered, including instruction rescheduling, priority re-arbitration and temporary response level adjustment.

7. The sightseeing vehicle anti-collision control method based on multi-modal radar according to claim 1, characterized in that: The process of converting the hierarchical response strategy into a PWM control signal specifically includes: According to the preset PWM signal conversion rules, the discrete response levels are mapped to corresponding PWM duty cycle and frequency parameters; The hardware timer and digital signal generation module work together. The timer accurately controls the PWM period according to the system clock period, and the digital signal generation module adjusts the duty cycle according to the mapping rules. Monitor the amplitude and frequency of the PWM waveform in real time, and use a closed-loop feedback mechanism to fine-tune the waveform to avoid waveform distortion caused by hardware jitter or electromagnetic interference; Set up an abnormality detection mechanism. When the PWM waveform exceeds the preset amplitude or frequency range, an error flag is triggered and the abnormal event log is recorded.

8. The sightseeing vehicle collision avoidance control method based on multi-modal radar according to claim 1, characterized in that: The process of extracting the voltage amplitude to complete relay drive signal adaptation and hardware compatibility detection and generate control instructions specifically includes: The PWM waveform is converted into a voltage signal sample through the analog signal acquisition interface, and the digital sampling is completed using a high-precision analog-to-digital converter; Synchronous sampling technology is used to ensure that the sampling time is strictly aligned with the PWM signal period; The sampled voltage amplitude data is processed by a digital filter, and a finite impulse response filter is used to eliminate high-frequency noise and transient interference; The processed voltage amplitude is converted into the level parameter of the relay drive signal through the voltage amplitude mapping algorithm; According to the relay drive voltage threshold, the voltage threshold is dynamically adjusted to ensure the reliability and stability of the relay drive.

9. The sightseeing vehicle collision avoidance control method based on multi-modal radar according to claim 1, characterized in that: Parse execution records to determine brake release conditions, generate adjustment instructions based on PID control, and configure closed-loop parameters through feedback. The expressions for achieving system parameter synchronization and closed-loop control include: Define the dynamic adjustment mechanism of the brake release trigger threshold: in, is the dynamic brake release threshold; 、 、 is the weight coefficient; is the basic threshold; The energy gradient of the signal for the kth execution; For the Environmental interference factors; is the total number of environmental interference factor categories; is an indicator of environmental interference intensity; Furthermore, we construct a logical synthesis of state determination: in, It is the brake release status sign; is the activation function; is the normalization coefficient; is the multiplication operator; is the index variable; For the Confidence weight of the historical execution record; Perform matching for history; The number of historical execution records, which is the upper limit of the index; based on With the speed recovery parameters, perform the PID control calculation and build the proportional-integral-derivative term: in, Output instruction for PID control; is the speed error; is the target speed limit; is the current vehicle speed; 、 、 : dynamic PID coefficient; is the environmental interference coefficient; is the integration time interval; Furthermore, the safety boundary constraints of the control instructions are: in, For final adjustment instructions; is the PID output change gradient; is the gradient change threshold; 、 To control the safety range of instructions; is the limiting function; is the L2 norm; Further, based on , generate the closed-loop feedback configuration and define the time domain mapping for parameter synchronization: in, is the actuator driving voltage after synchronization; is the time decay weight; is the attenuation coefficient; For the current moment; For the The execution timestamp of the history; The number of historical execution records.

10. A sightseeing vehicle collision avoidance control system based on multimodal radar, applied to the method according to any one of claims 1 to 9, characterized in that: include: Installation and calibration unit, used to obtain multi-modal radar installation parameters and complete sensor calibration; Sector limitation unit, used to set the horizontal field of view angle coverage of the millimeter wave radar; A threshold and strategy configuration unit, used to configure the collision risk determination threshold and the dynamic arbitration strategy table; Data acquisition unit, used to collect ranging and speed data from millimeter-wave radar and point cloud data from lidar; A time synchronization unit, used to align the timestamps of multimodal radars and perform spatial coordinate mapping; An event determination unit generates a collision risk level based on the fused target trajectory sequence; An arbitration unit, used to decide hierarchical control instructions according to a priority policy table; Speed ​​limit signal output unit, used to send deceleration instructions to the power controller; A brake signal output unit, used to send an emergency brake command to the brake actuator; An alarm management unit, used to trigger the sound and light alarm device; The log and parameter update unit is used to record the control instruction timestamp and update the policy table parameters.

Images