A multi-channel elastic wave road surface fingerprint mapping and matching positioning method and system

Abstract

This invention relates to the field of vehicle intelligent perception technology, specifically to a multi-channel elastic wave road surface fingerprint mapping and matching localization method and system. The method includes the following steps: acquiring multi-channel elastic wave signals generated by the interaction between a vehicle or mobile platform and a local road surface using a multi-channel elastic wave sensor module; amplifying, filtering, impedance matching, and anti-aliasing processing of the multi-channel elastic wave signals using an analog front-end module, and performing multi-channel synchronous or near-synchronous sampling of the processed signals using an ADC sampling module, forming multi-channel sampling data packets according to a preset sampling packet duration; and combining several consecutive multi-channel sampling data packets into a statistical window using an MCU processing module, extracting window-level time-domain features, frequency-domain features, and channel distribution features within the statistical window. This invention achieves local road surface relocation under vision-free and laser-free conditions, effectively solving the problem of strong dependence on the external environment in existing vision / laser solutions.

Description

Technical Field

[0001] This invention relates to the field of vehicle intelligent sensing technology, specifically to a multi-channel elastic wave road surface fingerprint mapping and matching positioning method and system. Background Technology

[0002] With the development of autonomous vehicles, mobile robots, intelligent inspection equipment, and special-purpose vehicles, the demand for local positioning of vehicles in complex roads, unpaved surfaces, underground parking lots, mining areas, factory areas, warehouse passages, tunnels, and areas with limited GNSS signals is constantly increasing. Existing vehicle positioning and mapping solutions mainly rely on sensors such as satellite positioning, visual sensors, lidar, millimeter-wave radar, IMU, or wheel speedometers.

[0003] LiDAR and vision sensors can acquire geometric or image information of the external environment and achieve mapping and localization through feature matching, point cloud registration, or visual relocalization. However, in scenarios with dust, rain, snow, strong light, weak light, smoke, obstruction, mud splashes, low texture, or cost constraints, vision and LiDAR solutions may suffer from reduced perception quality, higher hardware costs, greater computing power requirements, or complex installation and maintenance.

[0004] When a vehicle travels on an uneven road, its tires, suspension, frame, chassis, and body structure are subjected to periodic or transient excitations from the road surface. Different local road surfaces, such as potholes, bumps, gravel, cracks, road joints, speed bumps, rough asphalt, and gravel, will produce different mechanical responses in the vehicle structure. This response can be acquired by elastic wave sensors at a high sampling rate, and under multi-channel sensor deployment conditions, the differences in energy, spectrum, and propagation between different installation areas can be revealed.

[0005] In existing technologies, local mapping and matching localization techniques based on LiDAR mainly rely on the geometric features of the external environment. When vehicles are in dusty, rainy, snowy, foggy, low-texture roads, or scenes with repetitive local environmental geometry, the stability of point cloud features decreases, thus affecting the accuracy of local map matching. Vision-based relocalization techniques require relatively stable lighting conditions and rich visual textures, and are prone to failure in scenes with drastic lighting changes or scarce textures. Road surface recognition techniques based on IMU or vehicle vibration mainly collect low-frequency motion responses of the vehicle body or suspension. They typically have low sampling rates and limited channel numbers, and are not sensitive enough to high-frequency transient responses such as stone impacts, small cracks, road joints, and chassis scratches, making it difficult to form fine-grained road surface fingerprints. In addition, directly caching raw high-sampling-rate elastic wave signals will put a lot of pressure on MCU storage. Taking 8 channels and 10kHz sampling as an example, if raw elastic wave data is continuously cached for 800ms or longer, it will occupy a large amount of RAM and increase the overhead of subsequent feature calculations and model matching, which is not conducive to real-time operation on low-cost MCUs.

[0006] Meanwhile, a single short-time window only reflects transient responses and is easily affected by vehicle speed, load, tire condition, and single random impacts. If map matching is based solely on a single window, false matches are likely to occur. Traditional elastic wave collision detection methods typically output whether a collision occurred, the collision intensity, or the collision location, and cannot form a reusable map structure, nor can they perform fingerprint matching and localization when the vehicle passes through the same road segment again. In tunnels, underground parking lots, factory areas, mining areas, urban canyons, or indoor-outdoor transition areas, GNSS signals may be unavailable or have increased errors. Without local relocation methods that do not rely on external wireless signals and vision / laser, the continuity and robustness of the vehicle positioning system will decrease. Summary of the Invention

[0007] The purpose of this invention is to provide a multi-channel elastic wave road surface fingerprint mapping and matching positioning method and system to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A multi-channel elastic wave road surface fingerprint mapping and matching localization method and system includes the following steps: S1. Use a multi-channel elastic wave sensor module to collect multi-channel elastic wave signals generated by the interaction between the vehicle or mobile platform and the local road surface. S2. The multi-channel elastic wave signal is amplified, filtered, impedance matched and anti-aliasing processed using an analog front-end module, and the processed signal is sampled synchronously or approximately synchronously using an ADC sampling module to form a multi-channel sampling data packet according to a preset sampling packet duration. S3. Use the MCU processing module to combine several consecutive multi-channel sampling data packets into a statistical window, and extract window-level time-domain features, frequency-domain features and channel distribution features within the statistical window; S4. Use the MCU processing module to write the window-level features of multiple consecutive statistical windows into the circular buffer, and generate elastic wave fingerprint fragments based on the consecutive window-level features in the circular buffer. S5. During the mapping stage, the elastic wave fingerprint fragments are written into the storage module according to the driving sequence of the vehicle or mobile platform to form a local road fingerprint map including map nodes and topological connection relationships. S6. During the matching and positioning stage, the current elastic wave fingerprint fragment is generated in real time, the historical elastic wave fingerprint fragments of the candidate map nodes are read from the storage module, and the comprehensive similarity between the current elastic wave fingerprint fragment and the historical elastic wave fingerprint fragment is calculated. S7. Determine the current matching map node based on the comprehensive similarity, matching threshold, and map topology continuity constraints, and output the local location, map node number, matching confidence, and road surface status category through the communication module; when the comprehensive similarity does not meet the matching threshold, output unknown road segment, outside area of ​​map, or matching failure status. The method does not require cameras, image sensors, lidar, or laser ranging devices as necessary inputs during mapping and matching positioning.

[0009] Preferably, in step S4, the circular buffer is used to cache window-level features of 10 consecutive statistical windows to form an 800ms elastic wave fingerprint segment. When a new statistical window's window-level feature is written, the oldest window feature is overwritten or removed, and the complete historical original elastic wave waveform is not cached.

[0010] Preferably, in step S6, the comprehensive similarity includes at least two of the following: time-domain feature similarity, frequency-domain feature similarity, channel distribution feature similarity, and road surface state category consistency score; the comprehensive similarity calculation formula is:

[0011] in, For temporal feature similarity, For frequency domain feature similarity, For channel distribution feature similarity, The consistency score for road surface condition categories; , , , All are not less than 0, and .

[0012] Preferably, the road surface condition label and the road surface condition category adopt a unified closed label set, including smooth road surface, rough road surface, pothole road surface, raised road surface, cracked road surface, joint road surface, gravel road surface, speed bump road surface, wet and slippery road surface, gravel / unpaved road surface, chassis scraping or foreign object impact condition, and unknown condition.

[0013] A multi-channel elastic wave road surface fingerprint mapping and matching positioning system includes a multi-channel elastic wave sensor module, an analog front-end module, an ADC sampling module, an MCU processing module, a storage module, and a communication module. The multi-channel elastic wave sensor module is used to collect multi-channel elastic wave signals generated in the vehicle structure when a vehicle or mobile platform comes into contact with, impacts, rubs, or vibrates with a local road surface. The analog front-end module is used to amplify, filter, impedance match, and anti-aliasing the multi-channel elastic wave signal. The ADC sampling module is used to perform multi-channel synchronous or near-synchronous sampling of the signal processed by the analog front-end module, and to form multi-channel sampling data packets according to the preset sampling packet duration. The MCU processing module is used to receive the multi-channel sampling data packets, combine several consecutive sampling data packets into a statistical window, extract the window-level time-domain features, frequency-domain features and channel distribution features of the statistical window, write the window-level features of multiple consecutive statistical windows into a circular buffer and generate elastic wave fingerprint fragments. The storage module is used to store elastic wave fingerprint fragments, local road surface fingerprint maps, map nodes, topological relationships, road surface status labels, matching thresholds, and normalization parameters. The communication module is used to output map node number, local location, matching confidence, road surface condition category, off-map condition or matching failure status; In the mapping stage, the MCU processing module writes the elastic wave fingerprint fragments into the storage module to construct a local road surface fingerprint map. In the matching and positioning stage, it matches the real-time elastic wave fingerprint fragments with historical fingerprint fragments in the local road surface fingerprint map to output a local matching and positioning result. The system does not require cameras, image sensors, lidar, or laser ranging devices as necessary inputs during the mapping and matching and positioning processes.

[0014] Preferably, the multi-channel elastic wave sensor module includes at least two elastic wave sensors, which are arranged at at least one location in the vehicle chassis, frame, wheel arches, suspension connection area, left and right sides of the vehicle body, body structural components, or mobile robot load-bearing structure.

[0015] Preferably, the ADC sampling module samples at a sampling frequency of 10kHz, forming a multi-channel sampling data packet every 20ms, with each channel containing 200 sampling points in each sampling data packet; The MCU processing module combines four consecutive sampled data packets into an 80ms statistical window, and writes the window-level features of ten consecutive statistical windows into a circular buffer to form an 800ms elastic wave fingerprint segment.

[0016] Preferably, the window-level time-domain features extracted by the MCU processing module include at least three of the following: energy, root mean square, peak value, maximum gradient, kurtosis, skewness, and zero-crossing rate. Window-level frequency domain characteristics include at least three of the following: band energy, dominant frequency, spectral centroid, spectral entropy, and band energy ratio. Channel distribution characteristics include at least one of the following: channel energy ratio, main response channel, channel energy dispersion, and arrival time difference characteristics.

[0017] Preferably, the MCU processing module calculates the comprehensive similarity between the real-time elastic wave fingerprint segment and the candidate historical fingerprint segment during the matching and positioning stage. The comprehensive similarity includes time-domain feature similarity, frequency-domain feature similarity, channel distribution feature similarity, and road surface state category consistency score. The weight coefficients of each similarity component satisfy the non-negative normalization constraint. The MCU processing module calculates similarity using at least one of cosine similarity and dynamic time warping (DTW), and combines this with map topological continuity constraints to filter candidate map nodes.

[0018] Preferably, the local road surface fingerprint map includes multiple map nodes and the topological connection relationship between the nodes. Each map node includes at least one or more of the following: node number, elastic wave fingerprint segment, road surface status label, impact intensity or excitation intensity, confidence level, number of repeated observations, and timestamp. When a vehicle passes through the same road segment multiple times, the mean, variance, stability, and confidence of the corresponding map node are updated based on the elastic wave fingerprints collected multiple times, thus achieving map self-updating.

[0019] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention enables local road surface relocation under conditions without vision or laser, effectively solving the problem of strong dependence on the external environment in existing vision / laser solutions. This invention does not rely on external images or laser point clouds, but instead utilizes the multi-channel elastic wave response generated when a vehicle comes into contact with a local road surface to construct a local elastic wave fingerprint map, and achieves local relocation through real-time fingerprint matching. Compared with lidar and vision solutions, this invention is not affected by environmental factors such as lighting, texture, dust, rain, snow, smoke, and occlusion. In scenarios where GNSS is limited or vision conditions are poor, such as underground garages, tunnels, mining areas, factory areas, and warehouse passages, it can independently provide local relocation capabilities, significantly improving the environmental adaptability and robustness of the positioning system.

[0020] (2) This invention solves the problem that ordinary vibration signals are difficult to express high-frequency transient road surface excitations, and realizes fine-grained road surface feature characterization. This invention uses a multi-channel elastic wave sensor to collect high-frequency transient responses in the vehicle structure and extracts time-domain, frequency-domain, and channel distribution features, thereby characterizing fine-grained local road surface excitations such as potholes, bumps, cracks, joints, gravel, speed bumps, chassis scraping, or foreign object impacts. Compared with the low-frequency vibration signals of ordinary IMUs or suspension sensors, the multi-channel high-sampling-rate elastic wave signals of this invention can capture richer road surface microstructure information, forming a more distinguishable road surface fingerprint, and effectively improving the recognizability between different road conditions.

[0021] (3) This invention significantly reduces the storage and computational overhead of the MCU, solving the problem that high-sampling-rate elastic wave signals are difficult to process on low-resource MCUs for extended periods. This invention divides the 8-channel, 10kHz elastic wave signal into 20ms packets and 80ms windows, and extracts low-dimensional time-frequency features within each statistical window. The system only caches the features of the most recent 10 windows, without caching the original 800ms waveform, thereby significantly reducing the RAM usage of the MCU. Compared to the scheme of directly caching the original high-sampling-rate signal, this invention reduces the storage requirements by several orders of magnitude through window-level feature compression and a ring caching mechanism, enabling local fingerprint generation and matching to run in real time on a low-cost MCU, facilitating large-scale deployment and productization.

[0022] (4) This invention improves map matching stability and solves the problems of instability and mismatches caused by single-window elastic wave events. This invention combines multiple consecutive window-level features into an 800ms elastic wave fingerprint segment, enabling the system to use the continuous response pattern of local road surfaces for matching, reducing mismatches caused by a single random impact. At the same time, by combining map topology continuity constraints and continuous matching constraints, candidate nodes are filtered based on the matching nodes of the previous moment and the allowed advance / backward range, further reducing the probability of mismatches in a single frame. Compared with the single-window detection scheme, the sequence fingerprint matching mechanism of this invention can make full use of the spatiotemporal continuity features of local road surfaces, significantly improving matching accuracy and stability.

[0023] (5) This invention enables the construction of reusable local road surface fingerprint maps, solving the problem that ordinary road surface condition recognition or collision detection cannot form reusable maps. This invention not only identifies the current road surface condition, but also organizes continuous fingerprint segments into map nodes and establishes topological connections, so that road surface response features can be reused when a vehicle passes by again. Each map node stores information such as elastic wave fingerprint segments, road surface condition labels, impact intensity, confidence level, and number of repeated observations, and supports map self-updating after multiple passes, adapting to response changes caused by tire condition, load changes, and road surface aging. Compared with traditional collision detection or single road condition recognition, the local road surface fingerprint map constructed by this invention has the characteristics of being reusable, updatable, and matchable, providing basic support for long-term vehicle positioning.

[0024] (6) This invention improves the continuity and redundancy of the positioning system, solving the problem of insufficient positioning continuity of absolute positioning sensors such as GNSS in obstructed scenarios. In scenarios where GNSS is limited or unavailable, such as underground parking garages, tunnels, mining areas, factory areas, and warehouse passages, this invention can improve the continuity of the positioning system by outputting auxiliary local positioning results through local road surface elastic wave fingerprint maps. Simultaneously, this invention can supplement vision, laser, GNSS, IMU, or wheel speed positioning systems, providing local repositioning references when the main positioning sensor fails or its quality deteriorates, forming a multi-source fusion redundant positioning architecture, effectively improving the positioning reliability and safety of vehicles in complex environments. Attached Figure Description

[0025] Figure 1 This is a hardware connection block diagram of the multi-channel elastic wave road surface fingerprint mapping and matching positioning system of the present invention; Figure 2 This is a flowchart of the mapping and matching positioning method for the multi-channel elastic wave road surface fingerprint mapping and matching positioning method of the present invention; Figure 3 This is a schematic diagram of the elastic wave fingerprint fragment generation and ring buffering mechanism of the present invention. Detailed Implementation

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

[0027] Please see Figure 1 As shown, the multi-channel elastic wave road surface fingerprint mapping and matching positioning system of the present invention includes a multi-channel elastic wave sensor module, an analog front-end module, an ADC sampling module, an MCU processing module, a storage module, and a communication module connected in sequence, which completely covers the entire link from signal acquisition to positioning result output.

[0028] The multi-channel elastic wave sensor module consists of at least two, preferably eight, piezoelectric elastic wave sensors, which are installed in the vehicle chassis, frame, wheel arches, suspension connection points, left and right side skirts, etc., respectively. It can receive elastic wave signals propagating in the vehicle structure from road excitation in all directions; the sensors output analog elastic wave signals to the analog front-end module.

[0029] The analog front-end module includes a charge amplifier circuit, a bandpass filter circuit, an impedance matching network, and an anti-aliasing filter. These components sequentially amplify the amplitude, filter out noise, perform impedance matching, and apply anti-aliasing to the original analog signal, outputting a conditioned signal that meets the sampling requirements of the ADC.

[0030] The ADC sampling module supports 8-channel synchronous or near-synchronous sampling, with a preferred sampling frequency of 10kHz. It samples the conditioned analog signal according to a fixed period of 20ms, and each period forms a multi-channel sampling data packet which is sent to the MCU processing module.

[0031] The MCU processing module is the core computing unit of the system. It adopts an embedded microcontroller and is responsible for executing all algorithm logic, such as receiving sampling data packets, constructing statistical windows, extracting window features, managing ring buffers, generating elastic wave fingerprints, constructing local maps, calculating sequence matching, and controlling result output.

[0032] The storage module uses the MCU's internal Flash or external non-volatile memory to persistently store data such as local road fingerprint maps, map node indexes, normalized parameters, matching thresholds, and tag systems, and supports read and write access by the MCU.

[0033] The communication module supports communication methods such as CAN, UART, SPI, RS485, Ethernet or BLE. It is responsible for sending information such as positioning results, road conditions, and confidence levels to the vehicle's main control, navigation module or host computer. It can also receive mapping and matching control commands issued from outside.

[0034] The entire system can independently complete mapping and positioning functions without the need for external environmental sensing devices such as cameras and LiDAR, and has strong environmental adaptability and low cost advantages.

[0035] Please see Figure 2 As shown, the present invention provides a multi-channel elastic wave road surface fingerprint mapping and matching positioning method, which includes the following steps: Step S1: Multi-channel elastic wave signal acquisition: During vehicle operation, the tires interact with the road surface's irregularities, potholes, bumps, gravel, and seams, generating transient mechanical impacts that propagate through the vehicle's structure as elastic waves. Eight elastic wave sensors positioned at different locations on the vehicle simultaneously detect these vibrations, outputting eight channels of raw analog elastic wave signals. Different road surface structures excite elastic waves with varying energies, spectral compositions, and spatial propagation characteristics, forming the physical basis of road fingerprints.

[0036] Step S2, Signal Conditioning and Sampling Packetization: The analog front-end module sequentially performs charge amplification, bandpass filtering, impedance matching, and anti-aliasing filtering on the 8-channel raw analog signals to remove DC offset and high-frequency noise, adjusting the signal to the optimal amplitude range of the ADC sampling.

[0037] The ADC sampling module synchronously samples the 8-channel conditioning signal at a sampling frequency of 10kHz, packaging it into a sampling data packet every 20ms. Assume the vehicle is equipped with... The elastic wave sensor, the first The first channel in the The original sampled values ​​of each sampling point are:

[0038] in, This refers to the number of channels. In a preferred embodiment, The sampling frequency is 10kHz.

[0039] A sampling data packet is generated every 20ms, and each channel contains 100 packets.

[0040] in: The number of sampling points for each channel in each sampled data packet; The sampling frequency; This refers to the duration of the data packet sampling.

[0041] when , At that time

[0042] That is, each channel contains 200 sampling points in each 20ms data packet.

[0043] Step S3, Statistical Window Construction and Window-Level Feature Extraction After receiving the sampled data packets, the MCU processing module combines four consecutive sampled data packets into an 80ms statistical window, which serves as the basic unit for feature extraction. Each sampled data packet is combined into a statistical window.

[0044] in: For the duration of the statistical window; The number of data packets that make up a statistics window; Duration of a single sampled data packet.

[0045] In a preferred embodiment, , , then

[0046] Each statistical window contains, for each channel, the following:

[0047] when , At that time

[0048] That is, each channel contains 800 sampling points within each 80ms statistical window.

[0049] Before feature extraction, DC removal processing is first performed on the signal within each statistical window to eliminate interference from sensor zero drift and DC components.

[0050] in:

[0051] The first after removing DC Channel signal; These are the original sampled values; For the first The mean of the channel within the current statistical window; This represents the number of sampling points for each channel in the current statistics window.

[0052] Furthermore, the mean and standard deviation obtained from the training phase statistics can be used for standardization.

[0053] in: For the first One original feature; The first statistic obtained during the training phase The mean of each feature; The first statistic obtained during the training phase Standard deviation of each characteristic; To prevent division by zero of extremely small constants; These are the standardized features.

[0054] Furthermore, the mean and standard deviation of offline statistics can be used to standardize the features, thereby improving the consistency of features across different sensors and installation locations.

[0055] After preprocessing, time-domain and frequency-domain features are extracted for each channel within each statistical window, and the global channel distribution features are calculated. Extract at least three time-domain features for each channel within each statistical window.

[0056] (1) Window energy characteristics

[0057] in: For the first The first statistical window The energy of the channel; For the first Within the first statistical window Channel 1 One DC sampling value.

[0058] (2) Root Mean Square Characteristics

[0059] in: For the first The first statistical window The root mean square value of the channel.

[0060] (3) Peak characteristics

[0061] in: For the first The first statistical window The maximum absolute amplitude of the channel.

[0062] (4) Maximum gradient feature

[0063] in: For the first The first statistical window The maximum adjacent sample change of a channel.

[0064] In practice, time-domain features can be selected from at least three features such as energy, root mean square, peak value, maximum gradient, kurtosis, skewness, and zero-crossing rate.

[0065] (2) Frequency domain feature extraction Perform frequency domain analysis on each channel within each statistical window and extract at least three frequency domain features.

[0066] Let the first Channel 1 The discrete Fourier transform of each statistical window is as follows:

[0067] in: For the first The first statistical window The channel is in The spectrum of each frequency point; Frequency index; This is for the statistical window length.

[0068] (1) Frequency band energy characteristics

[0069] in: For the first The first statistical window The channel is in Energy within a frequency band; For the first The set of frequency points corresponding to each frequency band.

[0070] (2) Main frequency characteristics

[0071] in: For the first The first statistical window The channel's main frequency corresponds to the frequency point.

[0072] (3) Spectral centroid characteristics

[0073] in: The centroid of the spectrum; For the first The actual frequency corresponding to each frequency point; To prevent division by zero of extremely small constants.

[0074] (4) Frequency band energy ratio characteristics

[0075] in: For the first The proportion of energy in each frequency band to the total energy in the entire frequency band; This refers to the number of frequency bands.

[0076] In practice, frequency domain characteristics can be selected from at least three of the following: low-frequency band energy, mid-frequency band energy, high-frequency band energy, dominant frequency, spectral centroid, spectral entropy, and frequency band energy ratio.

[0077] (3) Extraction of channel distribution features Multi-channel elastic wave sensors can reflect the response differences at different installation locations, thus allowing for further calculation of channel distribution characteristics.

[0078] (1) Channel energy ratio

[0079] in: For the first The channel is in Energy percentage in each statistical window; This represents the total number of channels.

[0080] (2) Main response channel

[0081] in: For the first The main response channel in the statistics window.

[0082] (3) Channel energy dispersion

[0083] in:

[0084] For the first The degree of dispersion of multi-channel energy distribution within a statistical window; This represents the average energy across multiple channels.

[0085] The channel distribution characteristics can be used to distinguish different road excitation modes such as single-sided impact, double-sided bumps, chassis scraping, and continuous gravel impact.

[0086] Step S4, Circular Buffer and Elastic Wave Fingerprint Fragment Generation: Please see Figure 3 As shown, the MCU processing module uses a ring buffer mechanism to manage window-level features, without buffering the original sampled waveforms, but only buffering low-dimensional window-level feature vectors.

[0087] (1) Window-level feature vector construction For the A statistical window is created by concatenating the time-domain, frequency-domain, and channel distribution characteristics of all channels into a window-level elastic wave feature vector.

[0088] in: For the first Window-level elastic wave feature vectors corresponding to each statistical window; For window-level feature dimensions.

[0089] In one implementation, if each channel extracts 3 time-domain features and 3 frequency-domain features, and the number of channels is 8, then the dimension of the channel-level features alone is [value missing].

[0090] Optionally, global channel distribution characteristics such as channel energy ratio, main response channel, and channel energy dispersion can also be spliced.

[0091] In this embodiment, 3 time-domain features and 3 frequency-domain features are extracted for each channel, totaling 48 dimensions for 8 channels. In addition, the global channel distribution features result in a single-window feature dimension of approximately 50 dimensions, which is much lower than the original 800-point sampling dimension.

[0092] (2) Historical feature caching To adapt to real-time processing on the MCU side, this application does not buffer the complete original elastic wave signal for 800ms, but instead buffers the window-level feature vectors of multiple consecutive statistical windows.

[0093] Let the length of the history cache be... Then the first The historical features at each moment are cached as follows:

[0094] in: For the first Historical feature cache at the end of each statistical window; This represents the number of cached windows; For the first Window-level feature vectors of statistical windows.

[0095] In a preferred embodiment, Each statistical window is 80ms, so the historical feature cache corresponds to the elastic wave response of the most recent 800ms.

[0096] When new window features During generation, the MCU only... Write to the circular buffer and overwrite or remove the oldest window feature. It does not recalculate the historical window features, nor does it store the complete original sampled waveform.

[0097] (3) Generation of elastic wave fingerprint fragments Cache historical features The elastic wave fingerprint segment of the current local road section is

[0098] in: For the first Local elastic wave fingerprint fragments generated at each time point; This represents the number of statistical windows contained in the fingerprint fragment.

[0099] Elastic wave fingerprint fragments reflect the continuous structural response patterns generated by a vehicle passing over a local road surface over a recent period. Compared to single-window features, continuous fingerprint fragments can express the temporal evolution of road events, such as the attenuation after pothole impact, continuous high-frequency excitation of gravel roads, and the sequential responses of left and right lanes caused by speed bumps.

[0100] Step S5, Mapping Stage – Construction of Local Road Surface Fingerprint Map: When the system is working in mapping mode, the MCU processing module will write the generated elastic wave fingerprint fragments into the storage module in the order of vehicle travel. Each fingerprint fragment corresponds to a map node, and a topological connection relationship is established between adjacent nodes to finally form a local road fingerprint map.

[0101] During the mapping phase, elastic wave fingerprint fragments generated during vehicle movement are written into the map in chronological order.

[0102] in: A map of local elastic waves; For the first One map node; This represents the number of map nodes.

[0103] Each map node can be represented as:

[0104] in: For the first Elastic wave fingerprint fragments corresponding to each map node; This is the road surface status label corresponding to this node; This refers to the impact strength or road surface excitation strength corresponding to that node. This represents the confidence level or stability of the node. This represents the number of repeated observations for this node.

[0105] Establish topological connections between adjacent map nodes

[0106] This topological connection represents the driving sequence of vehicles during the mapping phase. If a vehicle passes through the same road segment multiple times, the fingerprint of the same map node can be updated.

[0107] in: For the updated map fingerprint; The map fingerprint before the update; This is the elastic wave fingerprint observed in this study; Update the weights for the map.

[0108] Step S6 is the matching and localization stage – fingerprint similarity calculation. When the system is operating in matching and positioning mode, the MCU generates the current elastic wave fingerprint segment in real time, following the sampling, windowing, feature extraction, and caching process that is completely consistent with the mapping phase. Then, the historical fingerprints of the candidate map nodes are read from the storage module. Calculate the overall similarity between the two.

[0109] The overall similarity is composed of temporal feature similarity. Frequency domain feature similarity Channel distribution feature similarity Consistency score with road surface condition category Weighted fusion is obtained as

[0110] in: For real-time fingerprinting With map fingerprint The overall similarity; For temporal feature similarity; Frequency domain feature similarity; Similarity of channel distribution features; The consistency score for road surface condition categories; α, β, γ, and δ are the weighting coefficients of each similarity component.

[0111] Where α, β, γ, and δ correspond to the weights of time-domain feature similarity, frequency-domain feature similarity, channel distribution feature similarity, and road surface state category consistency score, respectively, and satisfy the following constraints: α≥0, β≥0, γ≥0, δ≥0, and α+β+γ+δ=1.

[0112] In one implementation, the weights can be calibrated offline based on different vehicle platforms, road surface types, and sensor installation locations; in another implementation, the weights can be dynamically adjusted based on map node stability or real-time signal quality, but the adjusted weights still satisfy the above-mentioned non-negative normalization constraints.

[0113] Cosine similarity can be used to calculate the similarity of features of a single class.

[0114] in: , These are two feature vectors to be compared; To prevent division by zero of extremely small constants.

[0115] If changes in vehicle speed cause stretching or compression of the elastic wave fingerprint on the same road segment over time, DTW can be used to calculate the sequence distance.

[0116] in: For dynamic time-normalized distance; For time-aligned paths; The first in the real-time fingerprint sequence Each window feature; For the first in the map fingerprint sequence Each window feature; This represents the distance between two window features.

[0117] DTW distance can be converted into similarity.

[0118] Step S7 is the matching and localization output under topological constraints. To avoid single false matches, this application introduces map topology constraints on top of similarity matching. Let the map node matched in the previous time step be... The candidate node at the current moment is Then the candidate node should satisfy the following condition:

[0119] in: The number of nodes allowed to roll back; The number of nodes allowed to move forward.

[0120] In typical forward driving scenarios, a smaller setting can be used. and larger The matching position is continuously advanced along the driving direction to constrain the matching position.

[0121] The final matching node is

[0122] in: The current best matching map node number; The set of candidate nodes that satisfy the topological constraints.

[0123] When the highest similarity satisfies

[0124] If the match is successful, the map node will be output. .in: The threshold for a successful match.

[0125] If the highest similarity score is below the threshold, the output will be "unknown road segment" or "match failed".

[0126] Road surface condition recognition is as follows: In addition to matching and localization, this application can also input the current elastic wave fingerprint into a lightweight classification model, and output the road surface state category as follows:

[0127] in: A lightweight classification model deployed on the MCU or main control unit; This is the current real-time elastic wave fingerprint; This refers to the output road surface condition category.

[0128] The road surface condition categories adopt a unified closed label set, including smooth road surface, rough road surface, pothole road surface, raised road surface, cracked road surface, jointed road surface, gravel road surface, speed bump road surface, wet and slippery road surface, gravel / unpaved road surface, chassis scraping or foreign object impact condition, and unknown condition. In the same embodiment, the same label set is used in the mapping stage, the matching and positioning stage, and map node attributes; in different applications, the above categories can be merged, split, or expanded, but the expanded label set should remain consistent in the corresponding embodiment.

[0129] The classification results can be saved as map node attributes or used in map matching similarity calculations.

[0130] In addition, the MCU can run a lightweight road condition classification model simultaneously, inputting real-time elastic wave fingerprint fragments and outputting the current road condition category. The classification results can be used as auxiliary features for matching or output independently for road condition monitoring and vehicle control.

[0131] To adapt to the low computing power and low storage resources of the MCU, the system adopts the following optimization methods: Only window-level features are cached; the complete original historical sampling signals are not saved. For each new statistical window generated, only the features of the newly added window are calculated, and the features of the 9 historical windows are directly read from the circular cache without repeated calculation. During matching, calculations are performed preferentially within the candidate node range of topological constraints, without traversing the entire map; Multi-level indexes are created for map nodes to reduce the computational load of fine-grained matching through coarse screening; It supports feature fixed-point quantization and INT8 quantization, further reducing computational overhead.

[0132] Example 1: Mapping scenario for the first time opening of roads in the park This embodiment is applied to local road mapping for a park inspection vehicle. The target road section includes smooth asphalt sections, speed bump sections, road joint sections, gravel sections, and pothole sections. Eight piezoelectric elastic wave sensors are arranged at the front, rear, left, right, inner wheel arches, and suspension connection points of the inspection vehicle chassis. The sensor outputs are amplified and filtered by the analog front-end module and then connected to a 10kHz 8-channel ADC sampling module. The core calculation is handled by an STM32 series MCU, and the map data is stored in external Flash memory and communicates with the vehicle's main control unit via a CAN bus.

[0133] The vehicle travels at a constant speed along the target route for the first time. The ADC outputs an 8-channel sampling data packet every 20ms. The MCU splices four consecutive data packets into an 80ms statistical window and extracts three time-domain features (energy, root mean square, and peak value), three frequency-domain features (band energy, main frequency, and spectral centroid), and three channel distribution features (channel energy ratio, main response channel, and channel energy dispersion) window by window.

[0134] The MCU stores the features of the 10 most recent windows in a circular buffer, updating it with each new window and continuously generating 800ms elastic wave fingerprint segments. In mapping mode, the MCU writes each fingerprint segment to Flash in the driving sequence, generating corresponding map nodes, labeling road surface conditions and impact intensity, and establishing topological connections between nodes. After driving is complete, a local road surface fingerprint map covering the entire park road is generated, which can be used for matching and positioning during subsequent passages.

[0135] Example 2: Matching and Positioning Scenario for Repeated Passages in Underground Parking Garage This embodiment is applied to the local positioning of unmanned vehicles in underground parking garages, a scenario where GNSS signals are completely unavailable, lighting conditions are poor, and ordinary visual positioning is prone to failure. The vehicle has already pre-built an elastic wave fingerprint map of the garage passageway.

[0136] When a vehicle is driving in the garage, the system generates an 800ms elastic wave fingerprint segment in real time. The MCU reads map node data from Flash and determines the current candidate node range by combining the matching position and topological constraints of the previous moment. Then, it calculates the comprehensive similarity between the real-time fingerprint and the fingerprints of each candidate node, with a time domain weight of 0.3, a frequency domain weight of 0.4, a channel distribution weight of 0.2, and a state consistency weight of 0.1. Cosine similarity is used to calculate sequence similarity.

[0137] If the highest overall similarity score is higher than a preset threshold, the system outputs the currently matched map node number, the corresponding road segment location, and the matching confidence score, providing a local location reference for vehicle navigation. If the similarity score is lower than the threshold, the system outputs the off-map status, indicating that the vehicle has entered an unmapped area. The entire matching process is completely independent of cameras and LiDAR, and can still stably output local positioning results in low-light and dusty environments, effectively compensating for the lack of GNSS positioning capabilities.

[0138] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for multi-channel elastic wave road surface fingerprint mapping and matching localization, characterized in that, Includes the following steps: S1. Use a multi-channel elastic wave sensor module to collect multi-channel elastic wave signals generated by the interaction between the vehicle or mobile platform and the local road surface. S2. The multi-channel elastic wave signal is amplified, filtered, impedance matched and anti-aliasing processed using an analog front-end module, and the processed signal is sampled synchronously or approximately synchronously using an ADC sampling module to form a multi-channel sampling data packet according to a preset sampling packet duration. S3. Use the MCU processing module to combine several consecutive multi-channel sampling data packets into a statistical window, and extract window-level time-domain features, frequency-domain features and channel distribution features within the statistical window; S4. Use the MCU processing module to write the window-level features of multiple consecutive statistical windows into the circular buffer, and generate elastic wave fingerprint fragments based on the consecutive window-level features in the circular buffer. S5. During the mapping stage, the elastic wave fingerprint fragments are written into the storage module according to the driving sequence of the vehicle or mobile platform to form a local road fingerprint map including map nodes and topological connection relationships. S6. During the matching and positioning stage, the current elastic wave fingerprint fragment is generated in real time, the historical elastic wave fingerprint fragments of the candidate map nodes are read from the storage module, and the comprehensive similarity between the current elastic wave fingerprint fragment and the historical elastic wave fingerprint fragment is calculated. S7. Determine the current matching map node based on the comprehensive similarity, matching threshold, and map topology continuity constraints, and output the local location, map node number, matching confidence, and road surface status category through the communication module; when the comprehensive similarity does not meet the matching threshold, output unknown road segment, outside area of ​​map, or matching failure status. The method does not require cameras, image sensors, lidar, or laser ranging devices as necessary inputs during mapping and matching positioning.

2. The multi-channel elastic wave road surface fingerprint mapping and matching positioning method according to claim 1, characterized in that, In step S4, the circular buffer is used to cache window-level features of 10 consecutive statistical windows to form an 800ms elastic wave fingerprint segment. When a new statistical window's window-level feature is written, the oldest window feature is overwritten or removed, and the complete historical original elastic wave waveform is not cached.

3. The multi-channel elastic wave road surface fingerprint mapping and matching positioning method according to claim 1, characterized in that, In step S6, the comprehensive similarity includes at least two of the following: time-domain feature similarity, frequency-domain feature similarity, channel distribution feature similarity, and road surface state category consistency score; the comprehensive similarity calculation formula is:

4. Among them, For temporal feature similarity, For frequency domain feature similarity, For channel distribution feature similarity, The consistency score for road surface condition categories; , , , All are not less than 0, and .

5. The multi-channel elastic wave road surface fingerprint mapping and matching positioning method according to claim 1, characterized in that, The road surface condition labels and road surface condition categories adopt a unified closed label set, including smooth road surface, rough road surface, pothole road surface, raised road surface, cracked road surface, joint road surface, gravel road surface, speed bump road surface, wet and slippery road surface, gravel / unpaved road surface, chassis scraping or foreign object impact condition, and unknown condition.

6. A multi-channel elastic wave road surface fingerprint mapping and matching positioning system, characterized in that, It includes a multi-channel elastic wave sensor module, an analog front-end module, an ADC sampling module, an MCU processing module, a storage module, and a communication module. The multi-channel elastic wave sensor module is used to collect multi-channel elastic wave signals generated in the vehicle structure when a vehicle or mobile platform comes into contact with, impacts, rubs, or vibrates with a local road surface. The analog front-end module is used to amplify, filter, impedance match, and anti-aliasing the multi-channel elastic wave signal. The ADC sampling module is used to perform multi-channel synchronous or near-synchronous sampling of the signal processed by the analog front-end module, and to form multi-channel sampling data packets according to the preset sampling packet duration. The MCU processing module is used to receive the multi-channel sampling data packets, combine several consecutive sampling data packets into a statistical window, extract the window-level time-domain features, frequency-domain features and channel distribution features of the statistical window, write the window-level features of multiple consecutive statistical windows into a circular buffer and generate elastic wave fingerprint fragments. The storage module is used to store elastic wave fingerprint fragments, local road surface fingerprint maps, map nodes, topological relationships, road surface status labels, matching thresholds, and normalization parameters. The communication module is used to output map node number, local location, matching confidence, road surface condition category, off-map condition or matching failure status; In the mapping stage, the MCU processing module writes the elastic wave fingerprint fragments into the storage module to construct a local road surface fingerprint map. In the matching and positioning stage, it matches the real-time elastic wave fingerprint fragments with historical fingerprint fragments in the local road surface fingerprint map to output a local matching and positioning result. The system does not require cameras, image sensors, lidar, or laser ranging devices as necessary inputs during the mapping and matching and positioning processes.

7. The multi-channel elastic wave road surface fingerprint mapping and matching positioning system according to claim 5, characterized in that, The multi-channel elastic wave sensor module includes at least two elastic wave sensors, which are arranged at at least one of the following locations: vehicle chassis, frame, wheel arches, suspension connection area, left and right sides of the vehicle body, body structural components, or mobile robot load-bearing structure.

8. The multi-channel elastic wave road surface fingerprint mapping and matching positioning system according to claim 5, characterized in that, The ADC sampling module samples at a sampling frequency of 10kHz, forming a multi-channel sampling data packet every 20ms, with each channel containing 200 sampling points in each sampling data packet; The MCU processing module combines four consecutive sampled data packets into an 80ms statistical window, and writes the window-level features of ten consecutive statistical windows into a circular buffer to form an 800ms elastic wave fingerprint segment.

9. The multi-channel elastic wave road surface fingerprint mapping and matching positioning system according to claim 5, characterized in that, The window-level time-domain features extracted by the MCU processing module include at least three of the following: energy, root mean square, peak value, maximum gradient, kurtosis, skewness, and zero-crossing rate. Window-level frequency domain characteristics include at least three of the following: band energy, dominant frequency, spectral centroid, spectral entropy, and band energy ratio. Channel distribution characteristics include at least one of the following: channel energy ratio, main response channel, channel energy dispersion, and arrival time difference characteristics.

10. The multi-channel elastic wave road surface fingerprint mapping and matching positioning system according to claim 5, characterized in that, The MCU processing module calculates the comprehensive similarity between real-time elastic wave fingerprint segments and candidate historical fingerprint segments during the matching and positioning stage. The comprehensive similarity includes time-domain feature similarity, frequency-domain feature similarity, channel distribution feature similarity, and road surface state category consistency score. The weight coefficients of each similarity component satisfy the non-negative normalization constraint. The MCU processing module calculates similarity using at least one of cosine similarity and dynamic time warping (DTW), and combines this with map topological continuity constraints to filter candidate map nodes.

11. The multi-channel elastic wave road surface fingerprint mapping and matching positioning system according to claim 5, characterized in that, The local road surface fingerprint map includes multiple map nodes and the topological connection relationship between the nodes. Each map node includes at least one or more of the following: node number, elastic wave fingerprint segment, road surface status label, impact intensity or excitation intensity, confidence level, number of repeated observations, and timestamp. When a vehicle passes through the same road segment multiple times, the mean, variance, stability, and confidence of the corresponding map node are updated based on the elastic wave fingerprints collected multiple times, thus achieving map self-updating.

Images