UWB signal-based vehicle monitoring method and system, and vehicle

Abstract

The present application discloses a UWB signal-based vehicle monitoring method. The vehicle monitoring method comprises the following steps: signal transmission: transmitting a series of UWB signals by means of a UWB signal detector; signal reception: receiving, by means of the UWB signal detector, a UWB signal reflected back from a target object; signal analysis: analyzing, by means of a UWB signal processor, the received UWB signal to obtain monitoring data reflecting a state of the target object; and determination processing: on the basis of the monitoring data, generating a control signal by means of a system controller, and controlling a working state of a vehicle monitoring system by means of the control signal. The vehicle monitoring method of the present application enables effective perception and dynamic response to target objects around a vehicle body, improving real-time performance and accuracy of vehicle monitoring, and enhancing driving safety and system reliability.

Description

Vehicle monitoring methods, systems, and vehicles based on UWB signals Technical Field

[0001] This application relates to the field of vehicle monitoring technology, and in particular to a vehicle monitoring method, system and vehicle based on UWB signals. Background Technology

[0002] With the continuous development of automotive safety and monitoring technologies, in-vehicle video surveillance systems have been increasingly applied in the automotive field. Their core function is to record the surrounding environment of the vehicle through a recording system (including cameras and recorders) installed on the vehicle. These devices are typically kept continuously operational to capture and save surrounding video data in the event of a collision, scrape, or theft, thus aiding in incident tracking. However, continuous operation of the recording system leads to high power consumption. For gasoline-powered vehicles, this level of power consumption can significantly impact fuel consumption; for electric vehicles, it can severely shorten the driving range. Furthermore, because the monitoring system needs to record continuously, even when no abnormalities occur, it still generates a large amount of invalid monitoring data, wasting storage space and bandwidth resources.

[0003] In addition, UWB signals are prone to multipath interference during propagation due to reflection and refraction from obstacles, which leads to superposition distortion of received signals, resulting in decreased positioning accuracy and deviation in target feature extraction, directly affecting the accuracy of subsequent identification and decision-making.

[0004] Application content

[0005] The purpose of this application is to provide a low-power, high-efficiency vehicle monitoring method, system, and vehicle based on UWB signals.

[0006] To achieve the above objectives, this application provides a vehicle monitoring method based on UWB signals, comprising the following steps:

[0007] Signal transmission: A series of UWB signals are transmitted through a UWB signal detector.

[0008] Signal reception: The UWB signal reflected back after encountering the target object is received by the UWB signal detector.

[0009] Signal analysis: The received UWB signal is analyzed to obtain monitoring data reflecting the condition of the target object.

[0010] Judgment and processing: Generate control signals based on the monitoring data, and control the working status of the vehicle monitoring system through the control signals.

[0011] This application also provides a vehicle monitoring system based on UWB signals, which includes:

[0012] At least one UWB signal detector, mounted on the vehicle body, is used to transmit and receive UWB signals;

[0013] The UWB signal processor analyzes the received UWB signal to obtain monitoring data reflecting the status of the target object.

[0014] A vehicle video surveillance system, including at least one recording system;

[0015] The system controller is communicatively connected to the UWB signal processor and the recording system. The system controller is used to generate control signals based on the monitoring data to control the recording system to be in working or sleep mode.

[0016] Compared with the prior art, the vehicle monitoring method provided by the above-mentioned technical solution of this application can accurately acquire target monitoring data by transmitting, receiving and analyzing UWB signals, and then generate control signals to regulate the vehicle monitoring system after judgment and processing. This can achieve effective perception and dynamic response to targets around the vehicle, improve the real-time performance and accuracy of vehicle monitoring, and enhance driving safety and system reliability.

[0017] Compared with existing technologies, the vehicle monitoring system provided by the above-mentioned technical solution of this application includes a UWB signal detector installed on the vehicle body. The UWB signal detector transmits and receives UWB signals in real time. By processing the UWB signals reflected by the received target objects, monitoring data on the status of the detected target objects is obtained. Based on the monitoring data, corresponding control signals are generated to control the working state of the recording system. This allows the recording system on the vehicle to adaptively start or stop based on the status of nearby target objects. That is, when the target object has no impact on the vehicle, the recording system can be turned off, while when the target object may affect the vehicle, the recording system is automatically started. Thus, the above-mentioned vehicle monitoring system can significantly reduce system power consumption, improve vehicle range, and ensure that key image information can be captured at critical moments, achieving an organic combination of high efficiency, intelligence, and practicality. Attached Figure Description

[0018] Figure 1 is a schematic diagram of a vehicle equipped with a vehicle monitoring system in this application;

[0019] Figure 2 is a schematic diagram of the vehicle monitoring system in this application;

[0020] Figure 3 is a flowchart of the vehicle monitoring method in this application;

[0021] Figure 4 is a flowchart of the signal analysis steps in this application. Embodiments of the present invention

[0022] To explain in detail the technical content, structural features, objectives and effects of this application, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0023] In one embodiment of this application, a vehicle monitoring system based on UWB signals is disclosed for dynamic monitoring of the situation around a vehicle, as shown in Figures 1 and 2. The vehicle monitoring system includes at least one UWB signal detector 1, a UWB signal processor 2, a vehicle video monitoring system, and a system controller 3.

[0024] UWB signal detector 1 is mounted on the vehicle body. It is a radar sensor used to transmit and receive UWB signals. When the UWB signal emitted by UWB signal detector 1 is projected onto a target object around the vehicle body, an echo signal is returned and received by UWB signal detector 1.

[0025] The UWB signal processor 2 analyzes the received UWB signal (i.e., echo signal) to obtain monitoring data reflecting the condition of the target object.

[0026] The vehicle video surveillance system includes at least one recording system 4 installed inside the vehicle body. The recording system 4 is used to capture images of the area around the vehicle body through a camera and store the captured images.

[0027] The system controller 3 is communicatively connected to the UWB signal processor 2 and the recording system 4. The system controller 3 is used to generate control signals based on the monitoring data to control the recording system 4 to be in working or sleep mode.

[0028] Specifically, when the recording system 4 is in working condition, it captures and saves images of the area around the vehicle. When the recording system 4 is in sleep mode, it can maintain the power supply connection, consuming very little power, or it can disconnect the power supply to the recording system 4.

[0029] Based on the aforementioned vehicle monitoring system, when no anomalies are detected by the UWB signal detector 1, the recording system 4 remains in sleep mode, significantly reducing power consumption. Once an anomaly is detected approaching, the recording system 4 can be activated in real time, ensuring that critical images are captured at critical moments, reducing unnecessary recording, saving storage space, and improving monitoring efficiency. For electric vehicles, this greatly improves driving range; for gasoline vehicles, it reduces battery burden and provides effective real-time monitoring.

[0030] On the other hand, the monitoring data includes at least data reflecting the distance between the target and the vehicle body, as well as the state of the target. Specifically, the state of the target includes its motion state, such as whether it is stationary or moving. When it is moving, the monitoring data includes its speed and direction of movement relative to the vehicle body. In addition, the state of the target may also include the type of target, such as pedestrians, vehicles, animals, etc.

[0031] It should be noted that for UWB signal detector 1, the following different analysis methods can be used to obtain monitoring data.

[0032] Analysis Method 1 is as follows:

[0033] When analyzing changes in received UWB signals, due to the wide bandwidth characteristics of UWB signals, the received UWB signal can be decomposed into signals of multiple channels. Different target objects have different reflection and absorption characteristics of UWB signals, and these differences will be reflected in the signals of each channel. To accurately identify the type of moving object, the following methods can be used:

[0034] 1. Establish a channel feature database: By measuring and analyzing the UWB signal responses of various known targets, a corresponding channel feature database is established. These features include, but are not limited to, signal reflection intensity, attenuation degree, and multipath effects.

[0035] 2. Target object classification: For mechanically moving objects, such as remote-controlled cars, their surface material and shape may cause obvious reflection peaks in certain channels because their surfaces tend to reflect UWB signals at specific angles.

[0036] For a moving human body, due to its unique biological characteristics, such as water content and non-uniformity of muscle tissue, there will be a large signal attenuation in certain channels of UWB signals, which is different from the signal characteristics generated by mechanical objects.

[0037] 3. Signal Processing and Analysis: By comparing the real-time received UWB signal with channel characteristics in the database, the type of moving object can be identified. For example, if strong reflected signals are detected on some channels while the signal attenuates significantly on other channels, this may indicate that the target is a moving mechanical object.

[0038] 4. Multi-parameter comprehensive judgment: In addition to channel characteristics, other parameters, such as the multipath components of the signal, time difference of arrival, and signal phase, can be combined to achieve more accurate target identification.

[0039] This comprehensive analysis method can accurately identify the type of moving object, thereby improving the recognition capability and practicality of vehicle monitoring.

[0040] Analysis Method Two is as follows:

[0041] The target's movement characteristics are analyzed by monitoring continuous changes in the UWB signal. Since target movement causes corresponding fluctuations in the returned UWB signal over time, these continuous signal changes can be used to infer the target's movement characteristics. The specific analysis steps are as follows:

[0042] 1. Signal timing analysis: Collect a series of UWB signals received within a continuous time period and analyze their changing trends over time;

[0043] 2. Motion feature extraction: The movement of the target object will cause the following changes in the UWB signal:

[0044] Signal phase change: The movement of a target object will cause a change in the phase of the signal. By monitoring these phase changes, the movement trajectory of the target object can be inferred.

[0045] Changes in multipath effect: Moving the target object will cause changes in the multipath components of the signal. These changes reflect the position and velocity of the target object relative to the UWB signal detector 1.

[0046] Signal strength fluctuations: Changes in the size and speed of the target object can affect the signal reflection intensity and attenuation rate.

[0047] 3. Estimation of moving parameters:

[0048] Volume estimation: The volume of the target object can be estimated based on the reflection intensity and multipath effect of the received UWB signal.

[0049] Velocity calculation: By analyzing the phase change and multipath component movement speed of the received UWB signal, the movement speed of the target object can be calculated.

[0050] 4. Dynamic Model Establishment: By combining the volume and velocity information of the target object, a dynamic model of the target object can be established to further analyze the target object's movement trajectory and possible movement intentions.

[0051] This analytical approach not only allows us to identify the movement characteristics of targets, such as their size and speed, but also enables a deeper understanding of their dynamic behavior, thus providing vehicle-mounted monitoring with more accurate target tracking and evaluation capabilities.

[0052] On the other hand, to avoid false triggering caused by occasional interference signals, the system controller 3 generates a control signal based on at least two sets of monitoring data generated before and after the current period. Specifically, it takes the current monitoring data and another monitoring data from the adjacent previous period in the time series, performs an average calculation on these two monitoring data, and generates a control signal based on the average calculation result.

[0053] On the other hand, when the target enters the preset first warning area Q, if the duration of the target's stay in the first warning area Q exceeds the first time threshold, the control signal instructs the recording system 4 to start, so that the recording system 4 is in working state; in the first warning area Q, the distance between the target and the vehicle body is less than or equal to the first distance threshold.

[0054] In this embodiment, in the first warning area Q, the distance between the target object and the vehicle body is less than or equal to a first distance threshold, for example, the first distance threshold is 10 meters and the first time threshold is 5 seconds. Then, when the target object (such as a humanoid creature) is detected to enter within 10 meters of the vehicle body and stay for more than 5 seconds, the recording system 4 is activated.

[0055] Furthermore, when the target object remains in the first warning area Q for a duration exceeding the second time threshold while the distance between it and the vehicle remains unchanged, the control signal instructs the recording system 4 to shut down, putting the recording system 4 into a sleep state; the second time threshold is greater than the first time threshold. Specifically, when the first time threshold is 5 seconds, the second time threshold can be set to 60 seconds. In this embodiment, when the target object remains in the first warning area Q for a long time without moving, it is highly unlikely to affect the vehicle; therefore, it is unnecessary to keep the recording system 4 in working condition for an extended period, thus saving energy.

[0056] On the other hand, there are at least five UWB signal transmitters. Moreover, in order to achieve the ideal all-round monitoring effect, several UWB signal detectors 1 are respectively set around the vehicle body. As shown in Figure 1, the five UWB signal transmitters are respectively set on the left and right sides of the front of the vehicle, one on the left and right sides of the rear of the vehicle, and one in the middle of the roof.

[0057] On the other hand, the UWB signal detector 1 can be installed inside or outside the vehicle body.

[0058] In another preferred embodiment of the present invention, a vehicle is also disclosed, the vehicle including a vehicle body and a vehicle monitoring system as described in the above embodiments disposed on the vehicle body.

[0059] In another embodiment, this application also provides a vehicle monitoring method based on UWB signals, the method comprising the following steps:

[0060] S100, Signal Transmission: Transmits a series of UWB signals through a UWB signal detector.

[0061] The UWB signal emitted by the UWB signal detector is a series of extremely short pulse signals or modulated continuous waveforms, which have an extremely wide bandwidth. Due to the wide bandwidth characteristics of UWB signals, high time resolution can be provided, thereby enabling high-precision positioning by the UWB signal detector.

[0062] S200, Signal Reception: Receives the UWB signal reflected back after encountering the target object via a UWB signal detector.

[0063] The UWB signal emitted by the UWB signal detector propagates through space and is reflected back when it encounters a target. Due to the wide bandwidth characteristics of UWB signals, they can effectively distinguish signals on different paths even in complex multipath environments.

[0064] S300 Signal Analysis: The received UWB signal is analyzed by the UWB signal processor to obtain monitoring data reflecting the condition of the target object.

[0065] The monitoring data reflecting the status of the target object includes, but is not limited to, the distance between the target object and the vehicle body, the motion state of the target object, the azimuth angle of the target object relative to the vehicle body, and the type of the target object.

[0066] S400 Judgment and Processing: Based on the monitoring data, the system controller generates control signals and controls the working status of the vehicle monitoring system through the control signals.

[0067] The vehicle monitoring system's operating states include, but are not limited to, standby monitoring mode (when the target is far away or there is no target, the vehicle monitoring system is in low-power standby mode), Level 1 warning mode (when the target enters the first warning zone or slowly approaches the first warning zone, triggering a central control pop-up window, low-frequency prompts, etc.), Level 2 warning mode (when the target enters a second warning zone smaller than the first warning zone, is moving fast, or is pointing towards the vehicle, activating enhanced warning, automatic camera focusing, and simultaneous display of real-time images, etc.), and emergency response mode (when the target enters a collision risk range smaller than the second warning zone or experiences a sudden anomaly, issuing audible and visual alarms, and issuing deceleration requests, etc.).

[0068] Based on the above design steps, this application can accurately acquire target monitoring data by transmitting, receiving and analyzing UWB signals. After judgment and processing, control signals are generated to regulate the vehicle monitoring system, which can realize effective perception and dynamic response of targets around the vehicle, improve the real-time performance and accuracy of vehicle monitoring, and enhance driving safety and system reliability.

[0069] In another preferred embodiment of this application, a specific step design for S300 and signal analysis is also disclosed.

[0070] In a preferred embodiment, S300, signal analysis, can be achieved through the following steps:

[0071] S301, Distance and Velocity Information Acquisition: The received UWB signal is processed in a structured manner to extract and output the distance and velocity information of the target object.

[0072] Specifically, by performing structured processing on the UWB signal, basic quantization data is provided for subsequent steps, which also facilitates subsequent calculations / data processing.

[0073] S302, Azimuth information acquisition: Based on the signal characteristic differences of the received UWB signals, extract and output the azimuth information of the target object.

[0074] Specifically, azimuth information is extracted based on the characteristic differences of the same UWB signal (such as the phase difference received by multiple receiving modules in the same UWB signal detector), overcoming the limitation that a single receiving module cannot sense direction. By analyzing the signal characteristic differences, the azimuth of the target can be stably output in complex scenarios, improving the accuracy of azimuth measurement, supplementing the spatial positioning of the target with directional dimension information, and enhancing the comprehensive perception capability of the target's relative position.

[0075] S303. Target information fusion: Multi-source data fusion processing is performed on distance, velocity and azimuth information to obtain the spatial position and motion state of the target object.

[0076] Specifically, multi-source fusion of distance, velocity, and azimuth information can eliminate the limitations and errors of single-source data. Multi-source data fusion processing can improve data robustness, integrating scattered parameters into a unified spatial position (such as three-dimensional coordinates) and motion state (such as trajectory trend), more comprehensively reflecting the target dynamics, and providing complete and reliable feature input for subsequent higher-order type recognition functions.

[0077] S304. Target type recognition and output: Based on the fused target information, type recognition is performed, and the output is a comprehensive perception result including the target's spatial location, motion state and type.

[0078] Specifically, by identifying the type of target objects based on the fused target information, accurate classification can be achieved. The output comprehensive perception result contains all-round information about the target objects, which can provide more specific decision-making basis for vehicle monitoring systems, avoid the one-sidedness of judging solely by spatial position or motion state, and improve the adaptability and perception reliability of vehicle monitoring systems to complex scenarios.

[0079] In a preferred embodiment, step S301, the distance and speed information acquisition step, includes:

[0080] 1. When performing structured processing on the received UWB signal, a multi-dimensional signal model is first constructed. The multi-dimensional signal model includes signal representations that are associated with distance and velocity information (but not limited to associated distance and velocity information, it can also be associated with azimuth angle, phase difference and other information). Then, distance and velocity information are extracted based on the signal representation.

[0081] By constructing a multi-dimensional signal model that incorporates distance and velocity correlation representations, both are collaboratively extracted from the signal. Furthermore, the multi-dimensional signal model can map the intrinsic correlation between signal features and target object motion parameters, providing more relevant foundational data for subsequent fusion and accelerating computational efficiency.

[0082] 2. Before extracting distance and velocity information, interference suppression processing is performed on the received UWB signal, and the effective signal characteristics directly associated with the target are enhanced.

[0083] By suppressing interference in advance on UWB signals, irrelevant signals such as environmental noise and multipath reflections can be filtered out; the direct correlation features of the target can be enhanced, highlighting the effective components of the direct path signal. These steps reduce interference from invalid signals, resulting in cleaner extracted distance and velocity information, directly improving the accuracy and stability of parameter calculations.

[0084] Furthermore, the "construction of a multi-dimensional signal model" in the above preferred scheme can be achieved through the following steps:

[0085] Construct a CIR matrix that includes a fast time dimension and a slow time dimension; where the fast time dimension corresponds to the distance information of the target object, and the slow time dimension corresponds to the velocity information of the target object.

[0086] In the above steps, the CIR matrix captures the energy distribution of the signal over distance (such as the position of direct / multipath peaks) in the fast time dimension and records the trend of signal change over time (such as amplitude fluctuations caused by Doppler shift) in the slow time dimension. The distance and velocity information are associated and represented in the matrix, so that the signal features of the two can be extracted together. This avoids problems such as parameter asynchrony or data fragmentation caused by separate processing or staged processing, and improves the correlation and extraction efficiency of distance and velocity parameters.

[0087] Furthermore, the structured storage method of constructing a multi-dimensional signal model ensures that distance and velocity information are no longer isolated but are uniformly stored in the constructed matrix. Subsequent steps can directly extract the required parameters based on the row / column dimensions of the CIR matrix, avoiding data retrieval chaos. At the same time, the synchronous processing method ensures that the calculation of distance (based on the fast time dimension) and velocity (based on the slow time dimension) is based on the same set of signal data and the same time reference, eliminating parameter errors caused by time differences from the source and improving the consistency of distance and velocity.

[0088] More specifically, each row of the CIR matrix corresponds to all sampling points (TAPs) of a single pulse, and the value of each row corresponds to the complex signal value (including amplitude and phase) of this pulse. Among the complex signal values ​​in each row of the CIR matrix, the direct path signal reflected by the target object (the signal that best represents the true distance) will produce the largest amplitude fluctuation at a certain TAP. By finding this largest amplitude TAP, the distance of the target object can be calculated through its corresponding time delay, thus converting the fast time dimension (row) into distance information.

[0089] Meanwhile, each column of the CIR matrix corresponds to the set of TAP signal values ​​of all pulses at the same sampling time. Based on the Doppler effect, the movement of the target object will cause the signal phase to change. By analyzing the phase change law in the column direction of the matrix, the Doppler frequency shift can be calculated, and then the radial velocity of the target object can be calculated, thus converting the slow time dimension (column) into velocity information.

[0090] Furthermore, the "interference suppression processing of the received UWB signal" in the above preferred scheme can be achieved through the following steps:

[0091] Fourier transform (FFT) is performed on the time-domain signal of the CIR matrix to obtain the frequency-domain signal of the range-Doppler spectrum (RD spectrum); then frequency-domain equalization is used to compensate for the selective fading of the signal caused by multipath interference.

[0092] Furthermore, the "interference suppression processing of the received UWB signal" in the above preferred scheme can be achieved through the following steps:

[0093] Then, the frequency domain signal of the distance Doppler spectrum is converted back to the time domain signal of the CIR matrix by inverse Fourier transform (inverse FFT) to enhance the significance of the pulse peaks of the direct path of the UWB signal.

[0094] The essence of multipath interference is that signals of different frequencies attenuate to different degrees after passing through the channel. This difference is more intuitive in the frequency domain (e.g., some frequency bands are severely attenuated, while others are relatively normal). Multipath interference is difficult to separate in the time domain signal, but the frequency-selective fading it causes in the frequency domain signal follows a pattern.

[0095] Therefore, the above steps convert the time-domain signal of the CIR matrix into the frequency-domain signal of the range-Doppler spectrum pulse by pulse through FFT. This facilitates targeted compensation for selective fading caused by multipath interference in the frequency domain through frequency domain equalization, and accurately suppresses the attenuation imbalance of each frequency component.

[0096] Optionally, converting the frequency domain signal back to the time domain signal via inverse FFT can significantly enhance the amplitude and distinctiveness of the pulse peaks of the direct path, reduce the interference superposition of multipath reflection signals, and make the effective signal proportion higher when extracting distance and velocity information from the CIR matrix, thereby reducing the parameter extraction error caused by interference.

[0097] More specifically, the aforementioned "frequency domain equalization" can be achieved by designing a minimum mean square error (MMSE) equalizer, as follows:

[0098] By analyzing the attenuation law of the frequency domain signal, the compensation coefficient of the equalizer is calculated. For frequencies with severe attenuation, the compensation coefficient is greater than 1 (gain); for frequencies with normal attenuation, the compensation coefficient is close to 1 (no additional adjustment). Each frequency in the range-Doppler spectrum is multiplied by the corresponding compensation coefficient to obtain the equalized range-Doppler spectrum. At this time, the signal strength of each frequency has become consistent, the frequency imbalance caused by multipath interference is corrected, and the signal of all frequencies is restored to the ideal state when there is no multipath interference.

[0099] Furthermore, the "interference suppression processing of the received UWB signal" in the above preferred scheme can be achieved through the following steps:

[0100] The constant false alarm rate (CFAR) detection algorithm is used to process the time domain signal of the CIR matrix or the frequency domain signal of the range Doppler spectrum to identify and lock the pulse peak position of the direct path of the UWB signal, and use the pulse peak position of the direct path as the reference for extracting distance and velocity values.

[0101] Among them, the constant false alarm rate (CFAR) detection algorithm is a dynamic threshold detection algorithm. It dynamically adjusts the threshold (i.e., non-fixed threshold) based on the noise intensity around the detection point to determine whether the amplitude of the detection point exceeds the dynamic threshold. If it exceeds the threshold, it is determined to be a direct path peak; if it does not exceed the threshold, it is determined to be noise or multipath interference and is directly filtered out. Ultimately, it avoids false detection or missed detection caused by noise and locks the true direct path peak (i.e., the pulse peak of the direct path) in the signal.

[0102] Therefore, using a constant false alarm rate (CFAR) detection algorithm to process the time-domain signal of the CIR matrix or the frequency-domain signal of the range-Doppler spectrum can stably control the CFAR rate when noise fluctuates, avoid mistaking noise or weak multipath peaks as direct path peaks, accurately identify and lock the pulse peak position of the direct path, provide a reliable benchmark for the extraction of distance (fast time dimension) and velocity (slow time dimension) values, and significantly reduce the error of subsequent parameter calculations.

[0103] In a preferred embodiment, step S301, the distance and speed information acquisition step, further includes:

[0104] The extracted distance and velocity values ​​are optimized: based on the pulse peak position of the direct path, the parameter accuracy of distance and velocity values ​​is improved by resolution enhancement; multi-source observation data (i.e. distance and velocity information) are fused to reduce noise interference and output optimized distance and velocity values.

[0105] Specifically, the above steps use the pulse peak position of the direct path as a reliable benchmark, refine the quantification accuracy of distance and velocity parameters through resolution enhancement methods (such as reducing the minimum scale of distance measurement and increasing the velocity sampling interval), and fuse multi-source observation data (such as synchronous signals from different receiving modules and sampled values ​​of continuous time slices) to offset noise fluctuations of single data and reduce the impact of random interference. The final output of optimized distance and velocity values ​​is more accurate and more stable, providing a better dynamic parameter foundation for subsequent azimuth information fusion and target type identification.

[0106] Furthermore, the above-mentioned preferred solution's "improving the parameter accuracy of distance and velocity values ​​through resolution enhancement" can be achieved through the following steps:

[0107] 1. Delineate a local window centered on the pulse peak position of the direct path, and perform smooth interpolation algorithms on the real and imaginary components of the signal within the local window respectively. Generate subdivided sampling points by fitting the signal change trend.

[0108] In the above steps, defining a local window (e.g., a 6×6 grid) based on the pulse peak position along the direct path allows for focusing on the effective signal region, eliminating irrelevant data interference outside the local window, and reducing subsequent computational load. Performing smooth interpolation on the real and imaginary parts of the signal separately preserves the signal's phase and amplitude characteristics, avoiding information loss caused by single processing. The generated subdivided sampling points fill the gaps in the original sampling interval, significantly improving signal resolution and providing high-density, high-fidelity data support for subsequent precise pulse peak location optimization.

[0109] More specifically, the aforementioned "smooth interpolation" can employ the Spline interpolation method, performing Spline interpolation on the real and imaginary parts within a local window respectively. That is, drawing a continuous and smooth curve between the discrete points sampled by the hardware, and then inserting more virtual sampling points into the curve. After interpolation, the sampling point density is greatly improved, thereby achieving sub-pixel level resolution improvement through algorithmic point supplementation.

[0110] 2. Based on the constraints of the subdivided sampling points, the pulse peak optimization position of the direct path is repositioned, and the pulse peak optimization position of the direct path is used as the benchmark for extracting the distance and speed optimization values.

[0111] During the interpolation process, a large number of virtual sampling points will appear in the local window. However, the actual peak value of the direct path cannot be too far from the initial detection point found in the previous step. Therefore, a constraint range is set. Within the constraint range of the subdivided sampling points, the true peak value can be found while avoiding interference from pseudo-peaks outside the constraint range.

[0112] In the above steps, relying on the high-density data of subdivided sampling points, the true extreme points of the pulse peaks along the direct path can be captured more clearly. This effectively corrects the positioning deviation of the pulse peak positions along the direct path caused by large sampling intervals. Using the optimized position of the pulse peaks along the direct path as the benchmark for extracting optimized distance and velocity values ​​can directly eliminate parameter errors caused by inaccurate peak positioning, making the distance measurement scale more refined and the velocity calculation interval more precise, further improving the quantification accuracy and reliability of the two parameters. More specifically, the optimized distance and velocity values ​​can be obtained through the following preferred scheme.

[0113] In a preferred embodiment, step S301, the distance and speed information acquisition step, further includes:

[0114] 1. Distance calculation: using the formula Calculate the distance information of the target object; where d is the distance between the target object and the vehicle body, c is the speed of light, and Δt is the round-trip time of the UWB signal.

[0115] In the above steps, relying on the constant characteristic of the speed of light c and the measured value of the round-trip time Δt of the UWB signal, a linear correspondence between time and distance is established. The above calculation method avoids complex conversion errors. Combined with the high time resolution advantage of the UWB signal, it can accurately convert the subtle changes in the signal round-trip time Δt into distance values, ensuring centimeter-level accuracy in distance measurement and providing a direct and reliable quantitative basis for the spatial positioning of the target object.

[0116] 2. Speed ​​calculation: using the formula Calculate the velocity information of the target object; where v is the radial velocity of the target object, Δf is the frequency offset of the UWB signal, and λ is the wavelength of the UWB signal.

[0117] In the above steps, by utilizing the physical relationship between the frequency offset Δf of the UWB signal and the radial velocity v, and combining the fixed parameter of the signal wavelength λ, the frequency domain signal characteristics can be directly converted into velocity values. The above method does not require complex motion model derivation, can respond in real time to the frequency changes caused by the movement of the target object, improve the dynamic tracking capability of velocity measurement, and accurately reflect the rate at which the target object approaches or moves away.

[0118] More specifically, the frequency offset Δf is calculated using the formula Δf=f_d / f_c, where f_d is the Doppler frequency shift obtained by FFT transformation of the slow time dimension of the CIR matrix, and f_c is the carrier frequency.

[0119] It should be noted that the distance and speed information calculated and output by the above steps can correspond to both the unoptimized distance and speed values ​​and the optimized distance and speed values.

[0120] Furthermore, the "fusion of multi-source observation data to reduce noise interference" in the above-mentioned preferred scheme can be achieved through the following steps:

[0121] The observations from multiple receiving modules are fused using a weighted least squares method, and the result is obtained through the formula... The final distance information is calculated. The UWB signal detector includes multiple receiving modules arranged at preset intervals. w1 and w2 are inversely proportional to the signal-to-noise ratio difference between different receiving modules of the same UWB signal detector. d1 and d2 are the distance observation values ​​(i.e., distance calculation results) of different receiving modules of the same UWB signal detector, and d_final is the final distance information.

[0122] Specifically, the observations (i.e., measurement data) of a single receiving module are affected by noise, leading to errors in distance calculation. To address this, the above steps rely on multiple receiving modules (e.g., an antenna array) arranged at predetermined intervals within the UWB signal detector. By making the weights w1 and w2 inversely proportional to the signal-to-noise ratio of different receiving modules within the same UWB signal detector, the more accurate (smaller variance) observations d1 and d2 are given a larger proportion in the fusion process, reducing the interference from lower-precision data. This is then further processed using the formula... The weighted average effectively cancels out the random noise of each receiving module. The final output d_final not only significantly improves the distance measurement accuracy, but also reduces the error caused by the fluctuation of a single module, providing a more stable quantitative basis for subsequent target positioning.

[0123] It should be noted that the formula described in the above steps corresponds only to the fusion calculation method of the distance observation values ​​(d1 and d2) of two receiving modules of the same UWB signal detector. When the same UWB signal detector has three or more receiving modules, three or more distance observation values ​​will be generated accordingly. At this time, the third or more parameters (i.e., d3, w3; d4, w4; d5, w5...) are introduced, and the final distance information is obtained after fusion.

[0124] In addition, the above steps may also include a velocity information fusion processing step, which calculates the final velocity information by fusing velocity observations from multiple sources.

[0125] Furthermore, based on the above steps, observations from multiple UWB signal detectors located at different positions on the vehicle can be integrated to further improve the reliability and accuracy of the observations.

[0126] In a preferred embodiment, step S302, acquiring azimuth information, includes:

[0127] Based on the differences in signal characteristics among the same UWB signals received by multiple receiving modules, the azimuth angle value of the target object is output through stability optimization and accuracy improvement processing. The UWB signal detector includes multiple receiving modules arranged at preset intervals; each receiving module will produce different signal characteristics when receiving the same UWB signal.

[0128] The above steps utilize multiple receiver modules (e.g., antenna arrays) arranged at preset intervals. By leveraging the characteristic differences (such as phase difference and amplitude ratio) when receiving the same UWB signal, they provide a multi-dimensional reference for azimuth angle calculation, overcoming the limitation that a single receiver module cannot perceive direction. Simultaneously, through stability optimization to filter out transient interference and precision enhancement processing to refine angle quantization, the output azimuth angle value is not only more accurate but also more resistant to signal fluctuations. This supplements reliable directional dimension data for target spatial positioning, enhances the all-round perception capability of the target's relative position, and lays a precise angular information foundation for subsequent information fusion.

[0129] Furthermore, a phase difference is formed between the same UWB signals received by each receiving module, and the signal characteristic differences mentioned in the previous step are the phase difference. Based on this, the "azimuth angle value of the output target" in the above preferred scheme can be achieved through the following steps:

[0130] The Capon beamforming algorithm is used to search for the peak value of the spatial spectral function, and the phase difference corresponding to the peak value is taken as the phase difference value. The spatial spectral function is constructed based on the steering vector and the signal covariance matrix. The steering vector is the phase distribution template of the UWB signal in different receiving modules, and the signal covariance matrix is ​​calculated from the spacing between multiple receiving modules and the wavelength of the UWB signal.

[0131] In the steps described above, when constructing the spatial spectral function, the steering vector serves as a phase distribution template for the UWB signal incident on different receiving modules at a certain angle when it reaches the UWB signal detector. Combined with the covariance matrix calculated from the distance between the receiving modules and the signal wavelength, the spatial spectral function can accurately map the signal energy distribution at different angles, strengthening the physical correlation between the azimuth angle value and signal characteristics. Simultaneously, the Capon beamforming algorithm is used to search for spectral peaks, effectively suppressing sidelobe interference and accurately locking the angle corresponding to the direction of strongest energy as the azimuth angle value, laying a high-quality parameter foundation for subsequent processing.

[0132] More specifically, when multiple receiving modules receive UWB signals from the same target, the signal waveforms are similar, but there is a phase shift. The amount of phase shift is the phase difference Δ between the receiving modules. The above steps are used to record the phase difference Δ of all paired receiving modules. .

[0133] Furthermore, the "stability optimization and accuracy improvement processing" of the above preferred scheme can be achieved through the following steps:

[0134] The signal covariance matrix R is subjected to diagonal loading, which adds a diagonal perturbation term of a set strength to the signal covariance matrix R to improve the numerical stability of the phase difference.

[0135] The above steps add a diagonal perturbation term to the signal covariance matrix R, which can effectively improve the singularity problem that the matrix may cause due to insufficient samples or excessive noise, enhance the numerical stability of the matrix, avoid the phenomena such as spectral peak splitting and pseudo-peaks that occur in the Capon beamforming algorithm when the signal is weak or the interference is complex, keep the phase difference robust in noise fluctuations, reduce extreme value jumps, and provide a more reliable benchmark for subsequent optimization.

[0136] More specifically, the signal covariance matrix R needs to be calculated using sampled data from multiple receiver modules. However, in dynamic scenarios such as vehicle monitoring (e.g., rapid target movement and short sampling time), the amount of sampled data may be insufficient, causing the signal covariance matrix R to become a singular matrix (mathematically non-invertible). Capon beamforming requires calculating the inverse of the signal covariance matrix, and a singular matrix would cause the algorithm to crash, making it impossible to calculate the spatial spectral function P(θ). Therefore, the essence of the previous step is to ensure that the signal covariance matrix R is always invertible. Diagonal loading is equivalent to adding a stabilizing term, ensuring that the Capon beamforming algorithm can operate normally.

[0137] In an alternative approach, the initial azimuth angle can also be calculated directly using a spatial spectrum function, as follows:

[0138] At this point, the formula for the spatial spectral function is: Where a(θ) is the steering vector and R is the signal covariance matrix. It is the inverse of the signal covariance matrix, and θ is the azimuth angle to be calculated. The spatial spectrum function P(θ) is used to measure the degree of matching between the actual received signal and the steering vector a(θ). The higher the value of P(θ), the greater the probability that the signal is incident from the θ direction. Therefore, the θ value corresponding to the peak of the spatial spectrum is the initial value of the azimuth angle of the target object. The signal covariance matrix R is used to describe the correlation between the signals received by multiple receiving modules. aH(θ) is the conjugate transpose of a(θ).

[0139] Optionally, after obtaining the initial azimuth value through the spatial spectrum function, the historical angle and the current angle information can be combined to obtain more reliable and accurate angle information, i.e., the optimized azimuth value, through the Kalman filter algorithm.

[0140] The above steps use the current azimuth angle as the initial value to avoid redundant calculations and irrelevant interference in the global search. By using the Kalman filter algorithm for fine iteration within this local range, the mapping relationship between signal features and angles can be fitted using a probabilistic model to accurately locate the angle value with the smallest error. This significantly reduces the quantization error of the initial value, making the azimuth angle value closer to the true angle and improving the precision of angle measurement.

[0141] The initial azimuth angle mentioned above is a relative concept. Generally speaking, the initial azimuth angle value is obtained from the CIR matrix. In the iterative calculation of the spatial spectrum function, the value before each iteration can be called the initial azimuth angle value.

[0142] In a preferred embodiment, step S302, the azimuth information acquisition step, further includes:

[0143] Formula based on phase difference physical model The azimuth value is obtained by calculation. Here, θ is the azimuth value, and Δ... λ is the phase difference, d is the spacing between the receiving modules, and λ is the wavelength of the UWB signal.

[0144] The above steps calculate the azimuth angle value using the phase difference physical model formula, and then calculate the phase difference value Δ. Physical parameters such as the distance between receiving modules (d) and the signal wavelength (λ) are directly related to the azimuth angle (θ) of the target object.

[0145] In a preferred embodiment, step S303, the target information fusion step, includes:

[0146] 1. Time synchronization: Synchronize all observations to the same time reference; the observations include distance and velocity information and azimuth information observed by different UWB signal detectors.

[0147] The essence of the above time synchronization steps is to eliminate the target position deviation caused by time difference, and ensure that all subsequent data comparisons and fusions are based on the target state at the same moment, avoiding misjudgments caused by time asynchrony (for example, mistakenly thinking they are two different targets).

[0148] The above steps synchronize the distance, velocity, and azimuth observations from different UWB signal detectors to the same time reference, eliminating target position deviations caused by time differences between data, ensuring that all subsequent data comparisons and fusions are based on the target state at the same moment, and avoiding misjudgments caused by time asynchrony (such as mistakenly believing them to be two different targets).

[0149] Specifically, the same time reference can be the vehicle's system clock (such as the vehicle ECU clock). The sampling time of all UWB signal detectors is converted to this time reference. If the sampling time of a certain UWB signal detector does not coincide with the target time of the unified reference (data sampled asynchronously), linear interpolation is used to complete the data. Finally, the observation data of all modules are unified to the same time.

[0150] 2. Coordinate System 1: Transform the observations in the local coordinate system with the UWB signal detector as the origin to the global coordinate system with the rear axle center of the vehicle as the origin.

[0151] The above steps transform the observations in the local coordinate system (with the UWB signal detector as the origin) to the global coordinate system (with the rear axle center of the vehicle as the origin), unify the spatial reference, eliminate coordinate deviations caused by differences in the positions of different UWB signal detectors, enable multi-source data to be fused within the same spatial framework, and improve the overall spatial positioning of the target.

[0152] Specifically, when defining the global coordinate system, a Cartesian coordinate system (x, y) is established with the center of the vehicle's rear axle as the origin. For the observation value (d, θ) of each module, the coordinates of the UWB signal detector itself in the global coordinate system are calculated first, then the local Cartesian coordinates of the target object relative to the UWB signal detector are calculated, and then the local coordinates are converted into global coordinates. Finally, all the observation values ​​of the UWB signal detectors are converted into (x, y) coordinates in the global coordinate system, and the positions can be directly compared later.

[0153] 3. Uncertainty initialization: Assign a covariance matrix P to the observations; where the covariance matrix P is used to characterize the reliability of each observation.

[0154] Each UWB signal detector's observation data contains errors (e.g., the front module has smaller distance errors due to less obstruction, while the rear module has larger distance errors due to more multipath interference). Treating all observation data equally would result in large errors lowering the fusion accuracy. Therefore, it's necessary to quantify the reliability of each data point using initial uncertainty (higher reliability means lower uncertainty) to provide a basis for subsequent weighted fusion and prevent large errors from affecting the final result. The above steps quantify the error range of each observation data point using the covariance matrix P, allowing the fusion algorithm to process data differently based on reliability, reducing the interference weight of low-quality observations, providing a quantitative basis for error fusion of multi-source data, and improving the robustness of the results.

[0155] Specifically, the defined covariance matrix P is the initial uncertainty matrix of each observation. For each observation (x, y) of the UWB signal detector, its uncertainty is represented by an N×N covariance matrix P. Based on the hardware accuracy of the UWB signal detector and the actual scene calibration variance value, a corresponding covariance matrix P is assigned to each global coordinate (x, y). During subsequent fusion, the data weights will be adjusted according to the size of the covariance matrix P (the smaller P is, the greater the weight).

[0156] In a preferred embodiment, step S303, the target information fusion step, further includes:

[0157] 1. Redundancy detection: Set a spatial distance threshold; for any two observations from different UWB signal detectors, calculate the Euclidean distance between them in the global coordinate system; determine whether there is redundant correlation between the observations by comparing the magnitude of the Euclidean distance with the spatial distance threshold.

[0158] Specifically, if observations from two UWB signal detectors correspond to the same target, their spatial distance (i.e., Euclidean distance) in the global coordinate system will be very small. If they correspond to different targets or false targets, the distance will be very large. Therefore, the above step uses an Euclidean distance threshold to determine whether an observation is redundant (i.e., multiple observations of the same target). This step is used to initially screen potential observation groups of the same target, preparing for subsequent clustering and grouping, while excluding obvious false targets (e.g., data observed only by a single UWB signal detector and whose observation distance from other UWB signal detectors is far beyond the threshold).

[0159] The above steps, by setting a spatial distance threshold, calculate the Euclidean distance between observations from different UWB signal detectors in the global coordinate system and compare it with the threshold, can quickly identify redundant correlations between observations from multiple UWB signal detectors, avoid duplicate data interference caused by overlapping detection areas of the target object in subsequent fusion, and screen out candidate data with redundancy potential for clustering and grouping, ensuring the simplicity and correlation of data before fusion from the data source.

[0160] More specifically, for any two different UWB signal detectors, the Euclidean distance between the observed values ​​(x1, y1) and (x2, y2) is calculated as follows: First, calculate and square the difference between the coordinates of the two points along the x-axis; then calculate and square the difference between the coordinates of the two points along the y-axis; add these two squared results together; finally, take the square root of this sum. The formula is: .

[0161] 2. Clustering and grouping: A distance-based clustering algorithm is used to group observations that are close in location in the global coordinate system into the same redundant cluster; and a redundant cluster is determined to be a valid redundant cluster only when it contains at least two observations from different UWB signal detectors.

[0162] The above steps employ a distance-based clustering algorithm, which aggregates redundant observations from multiple UWB signal detectors into clusters based on spatial proximity in the global coordinate system. This means multiple observations of the same target are grouped into a single redundant cluster, counting only one target and avoiding duplicate counting. Furthermore, the redundant cluster must contain observations from at least two different detectors, ensuring that the data within each cluster represents valid multi-source observations of the real target (rather than noise or false points from a single detector). This provides a reliable multi-source redundant set for subsequent fusion, enhancing information density through the complementarity of observations from multiple detectors while filtering out meaningless single-point clusters (if a cluster contains only one observation from a UWB signal detector, it may be a false target and must be directly removed), thus improving the effectiveness of the fusion input data.

[0163] More specifically, the aforementioned "distance-based clustering algorithm" can automatically cluster observations that are close together into a redundant cluster without knowing the number of target objects in advance. It performs clustering, using coordinates in the global coordinate system as clustering features, and groups all observations according to the rule that the Euclidean distance is less than a set threshold. Each redundant cluster represents a potential target. By setting the validity conditions of the redundant clusters, multiple effective redundant clusters are finally obtained, and each cluster corresponds to all redundant observations of a real target object.

[0164] In a preferred embodiment, step S303, the target information fusion step, further includes:

[0165] Fusion Output: Based on the target motion model, the state prediction information of the target at the next moment is obtained; all observations in the effective redundant cluster and their corresponding observation noise covariance matrix are obtained; using the observations and their corresponding observation noise covariance matrix, the state prediction information is updated by the Kalman filter algorithm to obtain the fused state optimization information; the optimized state optimization information is output; the state prediction information and the state optimization information include the position and velocity information of the target.

[0166] Specifically, the target motion model in the above steps preferably adopts a constant velocity model (assuming the target object's velocity remains constant in a short time) to describe the change law of the target object's motion state at the prediction time; the observation noise covariance matrix (R1, R2, ...) is the covariance matrix corresponding to each observation value, used to quantify the noise level of each observation value (Z1, Z2, ...) within the effective redundant cluster; the update process of the Kalman filter algorithm specifically includes: calculating the state deviation based on the state prediction information and the observation values, correcting the deviation by combining the observation noise covariance matrix to optimize the target state, and thereby obtaining the state optimization information.

[0167] If the target object is in dynamic motion, the observation value at a single moment cannot accurately reflect its position and velocity information, and there are inherent error differences between different observation values ​​within the redundant cluster. Therefore, the above steps combine the target motion model and observation reliability to fuse and optimize the observation values. The Kalman filter algorithm can simultaneously achieve state prediction and observation update, ensuring that the prediction stage conforms to the actual motion law of the target object (such as uniform speed, acceleration, and deceleration), avoiding unfounded state estimation bias. Then, by utilizing the multi-source reliable observation values ​​of the effective redundant cluster, the deviation between the predicted value and the actual observation is accurately corrected, and the covariance corresponding to the observation value is updated to quantify the error range of the current state. Finally, the optimized state information of the target object is obtained, providing more quantitative basis for subsequent vehicle decisions (such as determining whether the target will collide).

[0168] In a preferred embodiment, S304, target type identification and output, includes:

[0169] 1. Feature extraction and fusion: Based on the target information obtained from signal analysis, extract high-dimensional features for classification; combine the high-dimensional features to form a high-dimensional feature vector.

[0170] Specifically, the target information obtained from the signal analysis includes both the observations from the UWB signal detector and the information obtained during the signal analysis process.

[0171] Including but not limited to:

[0172] a. The amplitude of the main peak of the CIR matrix waveform (the intensity of the strongest signal point), the width (the time span of the strongest signal point), the number of multipaths (the number of other small peaks besides the main peak), and the attenuation characteristics (the rate at which the signal intensity decreases after the main peak).

[0173] Due to differences in the material and shape of the target object, the reflected UWB signal patterns differ. For example, cars are made of metal, resulting in strong reflection and a large main peak amplitude; their large size leads to a large number of multipath signals (signal is reflected from different parts of the vehicle body); pedestrians are non-metallic (containing moisture), resulting in strong absorption and a small main peak amplitude; their small size leads to a small number of multipath signals.

[0174] b. Information such as distance, speed, and azimuth, as well as the calculated acceleration and its variation pattern (such as the gait cycle frequency of pedestrians and the uniform motion characteristics of vehicles).

[0175] Different targets have different motion characteristics, resulting in different speeds / accelerations and their variation patterns. For example, pedestrians have a gait cycle, leading to periodic small fluctuations in speed (such as one step faster and one step slower); cars typically move at a constant speed or accelerate uniformly, resulting in stable speeds without periodic fluctuations.

[0176] Based on a single observation of each target object, the specific values ​​of the two types of features (or more) mentioned above are calculated separately. All feature values ​​are then integrated into a high-dimensional feature vector. The combination of high-dimensional features can form a unique fingerprint (since a single feature cannot distinguish between different objects, for example, a large peak might indicate a car or a large guardrail). This fingerprint can then be directly fed into a deep learning model. Furthermore, the high-dimensional feature vector systematically integrates scattered target features, strengthens the correlation between features, provides input rich in discriminative information for subsequent target type identification, and improves the classification model's ability to distinguish between different target types.

[0177] 2. Database Construction: Collect multi-dimensional feature data of target objects of known types and form a sample database for model training; the multi-dimensional feature data includes high-dimensional feature vectors and corresponding target object type labels.

[0178] Specifically, deep learning models need a large number of samples of known types to learn to distinguish targets. Therefore, it is necessary to build a dedicated UWB feature database to provide learning materials for deep learning models. By comparing the correspondence between high-dimensional feature vectors and type labels, deep learning models gradually learn what feature combinations correspond to what target types (e.g., fast energy decay + broad spectrum corresponds to pedestrian targets).

[0179] Specifically, in this step, each data sample in the sample database must contain two parts: first, the input, which is the high-dimensional feature vector extracted in the previous step; and second, the label, which is the true type of the target object (such as a car or a pedestrian). The sample size needs to be large enough, as the accuracy of a deep learning model is positively correlated with the sample size. The more diverse the samples, the stronger the model's generalization ability (it can handle more unknown scenarios).

[0180] Therefore, the above steps construct a sample library by collecting multi-dimensional feature data (including high-dimensional feature vectors and type labels) of known target types, providing labeled training data for deep learning classification models. Multi-dimensional data covers the feature differences of different target types, while type labels establish a clear mapping between features and categories, enabling the model to learn stable classification rules. Simultaneously, the rich sample size enhances the model's adaptability to complex scenarios, laying a data foundation for the accuracy and generalization of target type recognition.

[0181] In one alternative approach, the target types can be divided into four categories (pedestrians, bicycles, cars and others), with each of the four different target types corresponding to a different high-dimensional feature vector.

[0182] In a preferred embodiment, S304, target type identification and output further includes:

[0183] Deep learning classification: A two-dimensional matrix is ​​formed by using CIR matrix data received by multiple receiving modules as the spatial dimension and CIR matrix data from multiple consecutive time slices as the temporal dimension. Combined with the sample database, a 2D-CNN (two-dimensional convolutional neural network) model is used to extract the spatiotemporal correlation features of the CIR matrix data. The mapping relationship between these spatiotemporal correlation features and target types is learned to achieve type classification output for the target object. The UWB signal detector includes multiple receiving modules (e.g., antenna arrays).

[0184] Specifically, the two-dimensional CIR matrix is ​​combined with the high-dimensional feature vector from the previous step, so that the model includes both the spatiotemporal features of the CIR matrix (correspondence between receiving modules and time) and global features (such as energy decay rate, velocity change, etc.). The samples in the sample database are divided into training set and test set. The training set data is fed into the 2D-CNN model to gradually optimize the correspondence between features and types. The test set is used to verify the model accuracy. If the accuracy is insufficient, the previous step is returned to supplement more samples.

[0185] More specifically, 2D-CNN (two-dimensional convolutional neural network) is used as the classification model. The core is to transform the feature data of UWB radio frequency signals into an image-like structure. By leveraging the advantage of 2D-CNN models in extracting spatial-temporal correlation features, the model can grasp the mapping rules between features and types. The 2D-CNN model extracts the temporal and spatial correlation features of CIR matrix data (such as pedestrian signals received by different receiving modules having similar attenuation patterns). Combined with high-dimensional feature vectors, accurate judgment of target type can be achieved.

[0186] Specifically, the sample data in the sample database is divided into a training set and a test set. The training set data is used to feed the deep learning model to gradually optimize the correspondence between target features and target types. The test set is used to verify the model's accuracy. If the accuracy is insufficient, the previous step is returned to supplement more samples.

[0187] In a preferred embodiment, S304, target type identification and output further includes:

[0188] Set a confidence threshold; output a confidence score for the classification results of different targets using the 2D-CNN model; if the confidence score of the classification result is higher than the confidence threshold, then the classification result is adopted; if the confidence score of the classification result is lower than the confidence threshold, then the classification result is marked as Unknown.

[0189] Specifically, the confidence threshold can be adjusted according to the scenario (a threshold that is too high will lead to missed detections, while a threshold that is too low will lead to false detections). If the confidence score of the classification result of a certain target type is greater than the set confidence threshold, the result is adopted and the target object is determined to be of the corresponding type. If the confidence score of the classification result of a certain target type is less than the set confidence threshold, it is not forcibly classified and is marked as Unknown. The vehicle will remain vigilant instead of ignoring it and will be prompted to conduct further observation (such as waiting for the target object to approach and then re-collecting features) to avoid errors caused by forcibly determining fuzzy classification and to ensure the reliability of the recognition results.

[0190] More specifically, the confidence score can preferably be the Softmax probability value, where the sum of the probabilities of all target types is 1. For example, for a certain target, the 2D-CNN model outputs "pedestrian probability = 0.92, car probability = 0.05, bicycle probability = 0.03", where 0.92 is the confidence score for the pedestrian type.

[0191] In another alternative approach, the following method can also be used to handle the reliability of classification results:

[0192] If the confidence score of the classification result of a certain target type is less than the set confidence threshold, then the target type with the highest probability value will be adopted as the result of this judgment.

[0193] In another preferred embodiment of the present invention, a vehicle is also disclosed that applies the vehicle monitoring method described in the above embodiments.

Claims

1. A vehicle monitoring method based on UWB signals, comprising the following steps: Signal transmission: A series of UWB signals are transmitted via a UWB signal detector; Signal reception: The UWB signal reflected back after encountering the target object is received by the UWB signal detector; Signal analysis: The received UWB signal is analyzed by a UWB signal processor to obtain monitoring data reflecting the condition of the target object; Judgment and processing: Based on the monitoring data, a control signal is generated by the system controller, and the working status of the vehicle monitoring system is controlled by the control signal.

2. The vehicle monitoring method according to claim 1, wherein The monitoring data includes any one or more of the following: The distance between the target object and the vehicle body; The motion state of the target object; The azimuth angle of the target object relative to the vehicle body; The type of the target object.

3. The vehicle monitoring method according to claim 1, wherein The signal analysis includes at least the following steps: Distance and velocity information acquisition: The received UWB signal is processed in a structured manner to extract and output the distance and velocity information of the target object; Azimuth information acquisition: Based on the signal feature differences of the same received UWB signal, the azimuth information of the target object is extracted and output; Target information fusion: Multi-source data fusion processing is performed on the distance and velocity information and the azimuth information to obtain the spatial position and motion state of the target object; Target type identification and output: Based on the fused target information, type identification is performed, and a comprehensive perception result including the spatial location, motion state and type of the target object is output.

4. The vehicle monitoring method according to claim 3, wherein The steps for obtaining distance and speed information include: When performing structured processing on the received UWB signal, a multi-dimensional signal model is first constructed, which includes a signal representation that is associated with distance and velocity information. Then, the distance and velocity information is extracted based on the signal representation. Before extracting the distance and speed information, interference suppression processing is performed on the received UWB signal, and the effective signal features directly associated with the target object are enhanced.

5. The vehicle monitoring method according to claim 4, wherein The construction of the multi-dimensional signal model includes: Construct a CIR matrix that includes a fast time dimension and a slow time dimension; wherein the fast time dimension corresponds to the distance information of the target object, and the slow time dimension corresponds to the velocity information of the target object.

6. The vehicle monitoring method according to claim 5, wherein The interference suppression processing performed on the received UWB signal includes: Perform a Fourier transform on the time-domain signal of the CIR matrix to obtain the range-Doppler spectrum; The selective fading of the UWB signal caused by multipath interference is compensated by frequency domain equalization.

7. The vehicle monitoring method according to claim 6, wherein The interference suppression processing performed on the received UWB signal further includes: The frequency domain signal of the distance-Doppler spectrum is converted back to the time domain signal of the CIR matrix by inverse Fourier transform in order to enhance the significance of the pulse peaks of the direct path of the UWB signal.

8. The vehicle monitoring method according to claim 6, wherein The interference suppression processing performed on the received UWB signal further includes: A constant false alarm rate (CFAR) detection algorithm is used to process the frequency domain signal of the range-Doppler spectrum, identify and lock the pulse peak position of the direct path of the UWB signal, and use the pulse peak position of the direct path as the reference for extracting distance and velocity values.

9. The vehicle monitoring method according to claim 7, wherein The interference suppression processing performed on the received UWB signal further includes: A constant false alarm rate (CFAR) detection algorithm is used to process the time-domain signal of the CIR matrix, identify and lock the pulse peak position of the direct path of the UWB signal, and use the pulse peak position of the direct path as the reference for extracting distance and velocity values.

10. The vehicle monitoring method according to any one of claims 8 or 9, wherein, The distance and speed information acquisition step further includes optimization processing, which includes: Based on the pulse peak position of the direct path, the parameter accuracy of the distance and velocity values ​​is improved by resolution enhancement methods; By fusing multi-source observation data to reduce noise interference, optimized distance and velocity values ​​are output.

11. The vehicle monitoring method according to claim 10, wherein The method of improving the accuracy of distance and velocity values ​​through resolution enhancement includes: A local window is defined centered on the pulse peak position of the direct path. A smooth interpolation algorithm is then applied to the real and imaginary parts of the signal within the local window to generate subdivided sampling points by fitting the signal change trend. Based on the constraints of the subdivided sampling points, the pulse peak optimization position of the direct path is repositioned, and the pulse peak optimization position of the direct path is used as the benchmark for extracting the distance and speed optimization values.

12. The vehicle monitoring method according to claim 11, wherein, The distance and speed information acquisition step further includes distance calculation and speed calculation, wherein the distance calculation and speed calculation include: By the formula Calculate the distance information of the target object; where d is the distance between the target object and the UWB signal detector, c is the speed of light, and Δt is the round-trip time of the UWB signal; By the formula Calculate the velocity information of the target object; where v is the radial velocity of the target object, Δf is the frequency offset of the UWB signal, and λ is the wavelength of the UWB signal.

13. The vehicle monitoring method according to claim 12, wherein, The fusion of multi-source observation data to reduce noise interference includes: The observation values of the plurality of receiving modules are fused by a weighted least square method, and the formula is Calculate the final distance information; wherein the UWB signal detector includes multiple receiving modules arranged at a preset interval, w1 and w2 are inversely proportional to the signal-to-noise ratio of different receiving modules of the same UWB signal detector, d1 and d2 are the distance observation values ​​of different receiving modules of the same UWB signal detector, and d_final is the final distance information.

14. The vehicle monitoring method according to claim 3, wherein, The steps for obtaining the azimuth information include: Based on the signal characteristic differences between the same UWB signal received by multiple receiving modules, the azimuth angle value of the target object is output through stability optimization and accuracy improvement processing; wherein the UWB signal detector includes multiple receiving modules arranged at a preset interval.

15. The vehicle monitoring method according to claim 14, wherein, The output of the azimuth angle value of the target object includes: The Capon beamforming algorithm is used to search for the peak value of the spatial spectrum function, and the phase difference corresponding to the peak value is used as the phase difference value. The spatial spectrum function is constructed based on the steering vector and the signal covariance matrix. The steering vector is the phase distribution template of the UWB signal in different receiving modules, and the signal covariance matrix is ​​calculated by the spacing between multiple receiving modules and the wavelength of the UWB signal.

16. The vehicle monitoring method according to claim 15, wherein, The stability optimization and accuracy improvement processes include: The signal covariance matrix is ​​subjected to diagonal loading processing, and a diagonal perturbation term of a set strength is added to the signal covariance matrix to improve the numerical stability of the phase difference value.

17. The vehicle monitoring method of claim 16, wherein, The acquisition of azimuth information also includes: Based on the phase difference physical model formula solving for the azimuth angle value; where θ is the azimuth angle value, Δ Let d be the phase difference value, d be the spacing between the receiving modules, and λ be the wavelength of the UWB signal.

18. The vehicle monitoring method according to claim 3, wherein, The target information fusion step includes time synchronization, coordinate system 1, and uncertainty initialization. The time synchronization, coordinate system 1, and uncertainty initialization include: All observations are synchronized to the same time base; wherein the observations include the distance and velocity information and the azimuth information observed by different UWB signal detectors; The observed values ​​in the local coordinate system with the UWB signal detector as the origin are transformed to the global coordinate system with the rear axle center of the vehicle as the origin; A covariance matrix is ​​assigned to the observations; wherein the covariance matrix is ​​used to characterize the reliability of each observation.

19. The vehicle monitoring method of claim 18, wherein, The target information fusion step further includes redundancy detection and clustering grouping, wherein the redundancy detection and the clustering grouping include: Set a spatial distance threshold; For any two observations from different UWB signal detectors, calculate the Euclidean distance between them in the global coordinate system; Determine the magnitude of the Euclidean distance and the spatial distance threshold, and determine whether there is redundant correlation between the observed values ​​based on the comparison result; A distance-based clustering algorithm is used to group the observations that are close in location in the global coordinate system into the same redundant cluster; Determine whether the redundant cluster contains at least two observations from different UWB signal detectors; if so, the redundant cluster is determined to be a valid redundant cluster.

20. The vehicle monitoring method according to claim 19, wherein, The target information fusion step further includes a fusion output, which includes: Based on the target motion model, the state prediction information of the target object at the next moment is obtained; Obtain all observations within the effective redundant cluster and their corresponding observation noise covariance matrix; Using the observed values ​​and the observed noise covariance matrix, the state prediction information is updated using the Kalman filter algorithm to obtain the fused state optimization information; The optimized state information is output; wherein the state prediction information and the state optimization information include the position and velocity information of the target object.

21. The vehicle monitoring method according to claim 3, wherein, The target type identification and output step includes feature extraction and fusion, and database construction, wherein the feature extraction and fusion and the database construction include: Based on the target information obtained from the signal analysis, high-dimensional features for classification are extracted. The high-dimensional features are combined to form a high-dimensional feature vector; Collect multi-dimensional feature data of the target object of known types, and form a sample database for model training from the multi-dimensional feature data; wherein the multi-dimensional feature data includes the high-dimensional feature vector and the type label corresponding to the target object.

22. The vehicle monitoring method of claim 21, wherein, The target type identification and output step further includes deep learning classification, which includes: A two-dimensional matrix is ​​formed by using CIR matrix data received by multiple receiving modules as the spatial dimension and CIR matrix data from multiple consecutive time slices as the temporal dimension; wherein the UWB signal detector includes multiple receiving modules. Using the sample database, the temporal and spatial correlation features of the CIR matrix data are extracted through a 2D-CNN model; By learning the mapping rules between the temporal and spatial correlation features and the target type, the type classification output of the target object can be achieved.

23. The vehicle monitoring method of claim 22, wherein, The target type identification and output step further includes classification result reliability processing, which includes: Set a confidence threshold; The 2D-CNN model outputs confidence scores for the classification results of different target objects. If the confidence score of the classification result is higher than the confidence threshold, then the classification result is adopted; If the confidence score of the classification result is lower than the confidence threshold, the classification result is marked as Unknown.

24. A vehicle monitoring system based on UWB signals, comprising: At least one UWB signal detector, mounted on the vehicle body, is used to transmit and receive UWB signals; The UWB signal processor analyzes the received UWB signal to obtain monitoring data reflecting the status of the target object. A vehicle video surveillance system, including at least one recording system; The system controller is communicatively connected to the UWB signal processor and the recording system. The system controller is used to generate control signals based on the monitoring data to control the recording system to be in working or sleep mode.

25. The vehicle monitoring system of claim 24, wherein, The monitoring data includes at least data reflecting the distance between the target and the vehicle body, as well as the state of the target.

26. The vehicle monitoring system of claim 24, wherein, When the recording system is in operation, it captures and saves images of the area around the vehicle; when the recording system is in sleep mode, its power supply is disconnected.

27. The vehicle monitoring system of claim 25, wherein, When the target object enters the first warning zone, if the duration of the target object's stay in the first warning zone exceeds a first time threshold, the control signal instructs the recording system to start, making the recording system operational; within the first warning zone, the distance between the target object and the vehicle body is less than or equal to a first distance threshold.

28. The vehicle monitoring system of claim 27, wherein, When the target remains in the first warning area for a duration exceeding the second time threshold and the distance between it and the vehicle remains unchanged, the control signal instructs the recording system to shut down, putting the recording system into a sleep state; the second time threshold is greater than the first time threshold.

29. The vehicle monitoring system of claim 24, wherein, The UWB signal detectors are at least five; and / or, a plurality of the UWB signal detectors are respectively arranged around the vehicle body; and / or, the UWB signal detectors may be arranged inside or outside the vehicle body.

30. The vehicle monitoring system of claim 24, wherein, The system controller generates the control signal based on at least two sets of monitoring data generated before and after the event.

31. A vehicle characterized by The vehicle monitoring method as described in any one of claims 1 to 23 is applied.

32. A vehicle characterized by Includes a vehicle body and a vehicle monitoring system as described in any one of claims 24 to 30, mounted on the vehicle body.

Images