Intelligent road condition recognition method, device and equipment for electric bicycle and storage medium

Abstract

The invention relates to an intelligent road condition recognition method, device and equipment for an electric bicycle and a storage medium, and the method comprises the following steps: carrying out the multi-frame image collection of a driving road surface through a binocular camera, and obtaining a road condition original image sequence; performing dynamic feature extraction on the road condition original image sequence to obtain a road surface dynamic feature map; performing multi-scale wavelet decomposition on the road surface dynamic characteristic spectrum to obtain a road condition characteristic component set; performing region division on the road condition feature component set through an adaptive threshold segmentation algorithm to obtain a road condition type feature region; carrying out boundary extraction and contour fitting on the road condition type feature region based on a Hough circle transformation technology to obtain road condition boundary mapping features; and performing spatial frequency domain analysis on the driving road surface based on the road condition boundary mapping features to obtain a road condition evaluation result, thereby solving the technical problem that detail features and dynamic change conditions of the road surface are difficult to accurately capture because most of the traditional technologies are based on single sensor data.

Description

Technical Field

[0001] The present invention relates to the technical field of electric bicycles, and particularly to an intelligent road condition recognition method, device, equipment and storage medium for an electric bicycle. Background Art

[0002] With the acceleration of the urbanization process and the enhancement of environmental protection awareness, electric bicycles, as a green and convenient means of transportation, have been widely used and developed globally. However, while providing convenience, electric bicycles also face challenges in driving safety, especially under complex and changeable road conditions. Traditional riding methods rely on the rider's experience judgment, which to a certain extent limits the safety and intelligence level of electric bicycles. Therefore, developing a method that can intelligently identify road conditions is crucial for improving the safety performance of electric bicycles.

[0003] The main existing problem is that current electric bicycles lack effective road condition perception capabilities. Although some high-end models are equipped with basic obstacle avoidance systems or navigation assistance functions, most of these technologies are based on single-sensor data (such as ultrasonic or monocular cameras), making it difficult to accurately capture the detailed features and dynamic changes of the road surface. In addition, due to the large differences in road conditions in different regions and seasons, fixed algorithm models are difficult to meet the accurate road condition recognition requirements in all environments. The existence of these problems has prompted researchers to explore more advanced technical means to improve the road condition recognition ability of electric bicycles.

[0004] To overcome the above challenges, researchers have begun to focus on how to use advanced visual sensing technologies and image processing algorithms to achieve intelligent recognition of road surface conditions. The application of technologies such as binocular cameras combined with optical flow field estimation, wavelet decomposition, and Hough transform provides new ideas for solving the road condition recognition problems faced by electric bicycles. Through the integrated application of this series of technologies, not only can high-precision road condition information be obtained in real time, but also different road condition types can be effectively distinguished, thereby providing more scientific and reasonable driving suggestions for riders and greatly improving the safety and comfort of electric bicycles. This innovative method represents an important direction for the future intelligent development of electric bicycles. Summary of the Invention

[0005] The main object of the present invention is to provide an intelligent road condition recognition method, device, equipment and storage medium for an electric bicycle, which solves the technical problem that most traditional technologies are based on single-sensor data and it is difficult to accurately capture the detailed features and dynamic changes of the road surface.

[0006] To achieve the above object, the present invention provides an intelligent road condition recognition method for an electric bicycle, wherein a binocular camera is provided on the electric bicycle, and the method includes the following steps: Collect multiple frames of images of the driving road surface through the binocular camera to obtain an original image sequence of the road conditions; Extract dynamic features from the original image sequence of the road conditions based on a preset optical flow field estimation algorithm to obtain a road surface dynamic feature map; Perform multi-scale wavelet decomposition on the road surface dynamic feature map to obtain a set of road condition feature components; Divide the set of road condition feature components through an adaptive threshold segmentation algorithm to obtain road condition type feature regions; Extract the boundary and fit the contour of the road condition type feature region based on the Hough circle transform technology to obtain a road condition boundary mapping feature; Perform spatial frequency domain analysis on the driving road surface based on the road condition boundary mapping feature to obtain a road condition evaluation result.

[0007] Further, the step of collecting multiple frames of images of the driving road surface through the binocular camera to obtain an original image sequence of the road conditions includes: Synchronously collect left and right images of the driving road surface through the binocular camera to obtain a pair of left and right images, and perform distortion correction on the pair of left and right images to obtain a corrected pair of left and right images; Perform exposure compensation on the corrected pair of left and right images to adjust the image brightness to obtain a pair of left and right images with balanced brightness, and perform image registration on the pair of left and right images with balanced brightness to accurately align the pair of left and right images to obtain a registered pair of left and right images; wherein, the registered pair of left and right images ensures the corresponding relationship of corresponding pixel points in space; Combine the registered pair of left and right images in chronological order to form an original image sequence of the road conditions.

[0008] Further, the step of extracting dynamic features from the original image sequence of the road conditions based on a preset optical flow field estimation algorithm to obtain a road surface dynamic feature map includes: Perform multi-directional gradient decomposition on the original image sequence of the road conditions based on a preset optical flow field estimation algorithm to obtain a group of image gradient vectors, and perform light intensity consistency constraint calculation on the group of image gradient vectors to obtain pixel light intensity change mapping data; wherein, the pixel light intensity change mapping data includes the light intensity change amount in the vertical direction and the light intensity change amount in the horizontal direction; Perform spatio-temporal correlation analysis on the original image sequence of the road conditions based on the pixel light intensity change mapping data to obtain a pixel motion vector field, and perform non-linear interpolation reconstruction on the pixel motion vector field to obtain a dense optical flow field description; Perform scene motion decomposition on the dense optical flow field description through a variational optimization method to obtain a road surface basic motion component, and perform local feature clustering on the road surface basic motion component to obtain a road condition motion feature block; Extract the high-order moment features of the road condition motion feature block to obtain a motion feature description sequence; Perform multi-dimensional feature fusion on the motion feature description sequence through tensor decomposition to obtain a road surface dynamic feature map, where the road surface dynamic feature map includes road surface deformation features and road surface vibration features.

[0009] Furthermore, perform multi-scale wavelet decomposition on the road surface dynamic feature map to obtain a set of road condition feature components, including: Perform orthogonal wavelet basis decomposition on the road surface dynamic feature map to obtain a multi-level decomposition coefficient matrix, and calculate the high-frequency sub-band energy of the multi-level decomposition coefficient matrix to obtain a road surface texture feature spectrum; where the road surface texture feature spectrum includes a roughness feature map and a flatness feature map; Perform band decomposition on the road surface texture feature spectrum through a preset wavelet packet transform technology to obtain a road surface frequency distribution feature, and perform subspace projection reconstruction on the road surface frequency distribution feature to obtain a set of road condition feature components; where the set of road condition feature components includes road surface geometric features, material features, and environmental illumination features.

[0010] Furthermore, perform regional division on the set of road condition feature components through an adaptive threshold segmentation technology to obtain road condition type feature regions, including: Perform multi-dimensional feature decomposition on the set of road condition feature components to obtain a road surface feature scale tensor, and perform local neighborhood analysis on the road surface feature scale tensor to obtain a road surface area gradient mapping matrix; Perform spatial correlation calculation on the road surface area gradient mapping matrix through a preset autocorrelation function to obtain a regional correlation feature map, and perform local variance estimation on the regional correlation feature map to obtain a regional heterogeneity description matrix; Through an adaptive threshold segmentation technology, perform adaptive threshold calculation on the regional correlation feature map based on the regional heterogeneity description matrix to obtain a multi-level threshold sequence, and perform region growing segmentation on the regional heterogeneity description matrix based on the multi-level threshold sequence to obtain an initial segmentation region set; Perform regional morphological processing on the initial segmentation region set to obtain a regional contour feature sequence, and perform boundary optimization reconstruction based on the regional contour feature sequence to obtain a reconstructed region boundary map, and perform regional topological relationship analysis on the reconstructed region boundary map to obtain a regional spatial association map; Perform regional label mapping on the regional spatial association map through a preset regional label mapping algorithm to obtain road condition type feature regions; where the road condition type feature regions include determined flat regions, concave-convex regions, and obstacle regions.

[0011] Further, the boundary extraction and contour fitting of the road condition type feature region are performed based on the Hough circle transformation technology to obtain the road condition boundary mapping feature, including: Perform polar coordinate transformation on the road condition type feature region through a preset Hough circle transformation technology to obtain a polar coordinate feature matrix, and perform circular arc feature detection on the polar coordinate feature matrix to obtain a road surface curvature feature set; wherein, the road surface curvature feature set includes the road surface circular arc curvature and the road surface boundary curvature; Perform edge refinement on the road surface curvature feature set through non-maximum suppression to obtain a refined edge feature map, and perform gradient direction statistics on the refined edge feature map to obtain an edge direction histogram; Perform circular contour voting on the refined edge feature map based on the edge direction histogram to obtain a circular contour candidate set, and perform boundary point extraction on the circular contour candidate set to obtain a boundary feature point sequence; Perform elliptical fitting transformation on the boundary feature point sequence to obtain a contour fitting parameter set, and perform boundary curve reconstruction based on the contour fitting parameter set to obtain a road condition boundary curve group; Perform shape feature extraction on the road condition boundary curve group through a preset geometric invariant moment technology to obtain a boundary shape descriptor, and perform feature clustering analysis on the boundary shape descriptor to obtain a road condition boundary mapping feature; wherein, the road condition boundary mapping feature includes a road surface contour mapping feature and a road surface boundary mapping feature.

[0012] Further, the spatial frequency domain analysis of the driving road surface is performed based on the road condition boundary mapping feature to obtain a road condition evaluation result, including: Perform bilateral spectrum decomposition on the road condition boundary mapping feature to obtain a frequency feature tensor, and perform band energy calculation on the frequency feature tensor to obtain a road surface vibration spectrum distribution map; wherein, the road surface vibration spectrum distribution map includes a vertical vibration spectrum and a horizontal vibration spectrum; Perform local feature extraction on the road surface vibration spectrum distribution map through joint time-frequency analysis to obtain a time-frequency feature sequence, and perform sub-band energy accumulation on the time-frequency feature sequence to obtain a road condition frequency feature matrix; Perform spectral peak detection on the time-frequency feature sequence based on the road condition frequency feature matrix to obtain a feature frequency set, and perform frequency clustering analysis on the feature frequency set to obtain a road condition spectrum pattern map; wherein, the road condition spectrum pattern map includes a steady frequency pattern and a mutation frequency pattern; Perform spatial phase correlation analysis on the road condition spectrum pattern map to obtain a phase correlation feature group, and perform spatial frequency reconstruction based on the phase correlation feature group to obtain a road surface frequency response map; Based on the road surface frequency response map, the road condition of the driving road surface is evaluated through a spatial spectral density estimation algorithm to obtain a road condition evaluation result.

[0013] The present invention also provides an intelligent road condition recognition device for an electric bicycle, characterized in that a binocular camera is provided on the electric bicycle, including: An acquisition module, configured to collect multiple frames of images of the driving road surface through the binocular camera to obtain an original road condition image sequence; An extraction module, configured to extract dynamic features of the original road condition image sequence based on a preset optical flow field estimation algorithm to obtain a road surface dynamic feature map; A decomposition module, configured to perform multi-scale wavelet decomposition on the road surface dynamic feature map to obtain a set of road condition feature components; A partitioning module, configured to partition the set of road condition feature components through an adaptive threshold segmentation algorithm to obtain a road condition type feature region; A fitting module, configured to extract the boundary and fit the contour of the road condition type feature region based on the Hough circle transform technique to obtain a road condition boundary mapping feature; An analysis module, configured to perform spatial frequency domain analysis on the driving road surface based on the road condition boundary mapping feature to obtain a road condition evaluation result.

[0014] The present invention also provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the method described in any one of the above are implemented.

[0015] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described in any one of the above are implemented.

[0016] An intelligent road condition recognition method for an electric bicycle provided by the present invention includes the following steps: collecting multiple frames of images of the driving road surface through the binocular camera to obtain an original road condition image sequence; extracting dynamic features from the original road condition image sequence to obtain a road surface dynamic feature map; performing multi-scale wavelet decomposition on the road surface dynamic feature map to obtain a set of road condition feature components; dividing regions of the set of road condition feature components through an adaptive threshold segmentation algorithm to obtain a road condition type feature region; extracting boundaries and fitting contours of the road condition type feature region based on the Hough circle transformation technology to obtain a road condition boundary mapping feature; performing spatial frequency domain analysis on the driving road surface based on the road condition boundary mapping feature to obtain a road condition evaluation result, solving the technical problem that most traditional technologies are based on single-sensor data and it is difficult to accurately capture the detailed features and dynamic changes of the road surface, and realizing generating a comprehensive and detailed road condition evaluation result through spatial frequency domain analysis of the driving road surface and combining various feature information obtained in the previous steps. This method can provide scientific and reasonable driving suggestions for cyclists, thus significantly improving the safety and comfort of electric bicycles. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a schematic diagram of the steps of an intelligent road condition recognition method for an electric bicycle in an embodiment of the present invention; Figure 2 is a structural block diagram of an intelligent road condition recognition device for an electric bicycle in an embodiment of the present invention; Figure 3 is a schematic structural diagram of a computer device in an embodiment of the present invention.

[0018] The implementation, functional features and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0020] As Figure 1 shown, Figure 1 is a schematic diagram of the steps of an intelligent road condition recognition method for an electric bicycle in an embodiment of the present invention; An intelligent road condition recognition method for an electric bicycle is provided in an embodiment of the present invention. A binocular camera is provided on the electric bicycle, and the method includes the following steps: Step S1, collecting multiple frames of images of the driving road surface through the binocular camera to obtain an original road condition image sequence.

[0021] Specifically, in the intelligent road condition recognition method of an electric bicycle, multi-frame image acquisition of the driving road surface by a binocular camera is a key step to achieve high-precision road condition perception. Specifically, the binocular camera consists of two lenses, simulating the principle of human binocular vision, and can capture the image information of the driving road surface from different angles simultaneously. These images not only contain the color and texture features of the road surface, but more importantly, by calculating the disparity between the two images, information about the road depth can be obtained, thereby constructing a three-dimensional road condition model. To ensure that the collected data is rich enough and representative, the system will continuously capture multiple frames of images to form a raw road condition image sequence. Each frame of image in this sequence contains the road conditions at different times. By analyzing the differences between these images, the changes in the road conditions can be dynamically tracked. For example, in a practical application scenario, assume that the electric bicycle is traveling on an urban road, and there is a bumpy section ahead, while other vehicles or pedestrians are passing by. At this time, the binocular camera will continuously capture images of this area at a certain frequency (such as 30 frames per second). Since the images captured by each lens are slightly different, the system can calculate the disparity map by comparing these two groups of images and further generate a depth map. This multi-perspective and multi-frame image acquisition method can not only accurately capture the undulation of the road surface, but also monitor the position and speed of moving objects in real time. In this way, even in a complex traffic environment, the electric bicycle can accurately identify potential risks and make timely responses, such as decelerating or detouring, based on these detailed raw road condition image sequences, combined with subsequent algorithms such as optical flow field estimation, thereby effectively improving the safety and comfort of riding. Therefore, this technical means is crucial for improving the intelligent level of electric bicycles. In this way, the system can provide more scientific and reasonable driving suggestions for riders, significantly improving the travel experience.

[0022] Step S2: Based on a preset optical flow field estimation algorithm, perform dynamic feature extraction on the raw road condition image sequence to obtain a road surface dynamic feature map.

[0023] Specifically, in the intelligent road condition recognition method of an electric bicycle, dynamically extracting features from the original road condition image sequence based on a preset optical flow field estimation algorithm is one of the key steps to achieve accurate road condition analysis. Specifically, the optical flow field estimation algorithm estimates the movement of pixel points in consecutive frame images to capture the dynamic changes of the road surface and its surrounding environment. First, the system uses the multi-frame image sequence obtained by the binocular camera as input, and these images contain detailed information about the road surface and the surrounding environment. Then, the algorithm generates an optical flow field that describes the movement trend of the entire scene by calculating the displacement vectors of each pixel point between adjacent frames. This optical flow field not only reflects the movement of objects in space but also reveals the relative movement speed and direction of different regions, thereby helping the system identify potential dynamic features, such as the movement of pedestrians and vehicles or changes in road surface obstacles. For example, in a practical application scenario, assume that an electric bicycle is traveling on an urban road, and there is a bumpy section ahead, and at the same time, there are other vehicles or pedestrians passing by. At this time, the optical flow field estimation algorithm will process the multi-frame images continuously captured by the binocular camera and calculate the position offset of each pixel point in each frame relative to the previous frame. In this way, the algorithm can identify the movement trajectories of pedestrians and vehicles and distinguish between the static road surface and dynamic traffic participants. In addition, the algorithm can also detect minor changes in the road surface, such as newly emerged small potholes or scattered stones. Finally, these dynamic features will be integrated into a road surface dynamic feature map, which not only contains the movement information of various objects but also shows the mutual relationship and influence between them. Based on this map, subsequent steps such as multi-scale wavelet decomposition can further analyze these dynamic features, provide more accurate road condition assessment and warning information for the rider, and thus significantly improve the safety and intelligence level of the electric bicycle. This process fully demonstrates the powerful ability of the optical flow field estimation algorithm in dynamic feature extraction and its important application value in intelligent road condition recognition.

[0024] Step S3: Perform multi-scale wavelet decomposition on the road surface dynamic feature map to obtain a set of road condition feature components.

[0025] Specifically, in the intelligent road condition recognition method for electric bicycles, performing multi-scale wavelet decomposition on the road surface dynamic feature map to obtain a set of road condition feature components is an important step for further refining the road condition information processing. Specifically, this process starts with using the road surface dynamic feature map generated in the previous step as the input and applying multi-scale wavelet decomposition technology to deeply analyze and extract the detailed features therein. Wavelet decomposition is a mathematical transformation method that can decompose signals or images at different scales, thereby effectively capturing local changes and global trends in the data. In this process, first, a suitable wavelet basis function is selected, and then a series of filtering operations are performed on the road surface dynamic feature map to separate road condition features with different frequency components. These features not only include macroscopic morphological information of the road surface, such as slopes and curves, but also can reflect dynamic changes at the microscopic level, such as potholes, stones, and the movement trajectories of pedestrians. For example, in an urban traffic environment, when an electric bicycle travels to a complex intersection, the road conditions are complex and constantly changing. At this time, by performing multi-scale wavelet decomposition on the road surface dynamic feature map, the system can accurately identify and distinguish obstacles of different sizes and shapes and their movement directions, and at the same time detect road unevenness. For example, it can identify an emergency deceleration caused by a vehicle suddenly braking ahead, or discover newly emerged small obstacles on the road surface, and convert this information into a set of road condition feature components at different scales. The advantage of doing this is that it can not only understand the traffic conditions of the entire section from a macroscopic perspective, but also carefully grasp every factor that may affect riding safety at the microscopic level. Finally, these sets of road condition feature components obtained through multi-scale wavelet decomposition provide more accurate data support for subsequent risk assessment and early warning, enabling electric bicycles to achieve more intelligent and safer navigation in complex urban environments. This process fully demonstrates the importance of multi-scale wavelet decomposition technology in improving the accuracy of road condition recognition and its potential value in intelligent transportation systems.

[0026] Step S4, perform regional division on the set of road condition feature components through an adaptive threshold segmentation algorithm to obtain road condition type feature regions.

[0027] Specifically, in the intelligent road condition recognition method for electric bicycles, dividing the road condition feature component set through the adaptive threshold segmentation algorithm to obtain the road condition type feature region is an important step in achieving accurate road condition classification. Specifically, this process starts with taking the road condition feature component set obtained from the multi-scale wavelet decomposition in the previous step as the input, using the adaptive threshold segmentation algorithm to automatically determine the optimal segmentation threshold, and accordingly dividing different regions in the image. The adaptive threshold segmentation algorithm can dynamically adjust the threshold according to the local features of the image, thereby effectively handling the influence of factors such as illumination changes, shadows, and complex backgrounds. First, the algorithm calculates the gray value or feature intensity of each pixel point and generates an initial threshold distribution map based on these values. Then, by analyzing the differences and similarities between adjacent pixel points, the algorithm continuously adjusts the threshold until the optimal segmentation scheme is found. For example, in an urban traffic environment, when an electric bicycle travels to a complex intersection, the road surface conditions are complex and constantly changing, including different road surface materials, colors, and the presence of various obstacles. At this time, by applying the adaptive threshold segmentation algorithm to the road condition feature component set, the system can accurately divide the image into multiple regions with different characteristics. For example, it can distinguish flat asphalt roads, damaged areas with cracks or potholes, sidewalk boundaries, and regions where dynamic objects such as pedestrians and vehicles are located. Suppose there is a severely damaged section ahead and there are pedestrians crossing the road at the same time. The adaptive threshold segmentation algorithm can automatically adjust the threshold according to the feature intensity and texture information of different regions to accurately segment these regions. This not only helps to identify specific road condition types, such as flat roads, pothole areas, or obstacles, but also further distinguishes the specific positions and movement trajectories of dynamic objects. Finally, the road condition type feature region obtained through this series of operations provides a solid foundation for subsequent boundary extraction and contour fitting, enabling the electric bicycle to achieve more accurate road condition assessment and safe navigation in a complex urban environment. This process demonstrates the flexibility and accuracy of the adaptive threshold segmentation algorithm in processing complex road condition data, which is crucial for improving the intelligent level of electric bicycles.

[0028] Step S5: Based on the Hough circle transform technology, perform boundary extraction and contour fitting on the road condition type feature region to obtain the road condition boundary mapping feature.

[0029] Specifically, based on the Hough circle transform technology, the boundary of the road condition type feature area is extracted and the contour is fitted to obtain the road condition boundary mapping feature. This process is achieved by identifying circular structures in the image and is particularly suitable for processing road condition information with obvious circular features. First, the Hough circle transform technology converts the original image space into the parameter space. In this process, each pixel point corresponds to a possible combination of the center of the circle and the radius. Then, the algorithm searches for the points with the highest cumulative value in the parameter space according to these combinations. These points with high cumulative values represent the positions where circles are most likely to appear in the image. Therefore, for the road condition type feature area obtained by the adaptive threshold segmentation algorithm, the Hough circle transform technology can accurately detect the circular elements contained therein and further extract their boundaries and contours. For example, during the process of an electric bicycle traveling on an urban road, it may encounter objects with circular characteristics such as manhole covers and traffic signs. When the vehicle approaches these objects, the system can accurately identify and extract the boundaries and contours of these circular objects by using the Hough circle transform technology. Suppose there is a circular manhole cover slightly protruding from the ground ahead. Through the Hough circle transform technology, the system can not only locate the position of the manhole cover but also obtain its specific size and shape characteristics. In this way, even under complex lighting conditions or in the presence of interfering objects, it can ensure accurate identification of the boundary of the manhole cover, providing key data support for subsequent road condition assessment. In addition, through the accurate extraction of the manhole cover boundary, the system can also analyze its wear degree or whether it has shifted, etc., so as to provide more secure navigation suggestions for the rider. In summary, the application of the Hough circle transform technology not only improves the recognition accuracy of specific road condition features but also lays a solid foundation for the overall performance improvement of the intelligent road condition recognition system. This enables the electric bicycle to more intelligently cope with various road condition challenges in the complex and changeable urban environment, ensuring the safety and comfort of riding.

[0030] Step S6: Based on the road condition boundary mapping feature, perform spatial frequency domain analysis on the driving road surface to obtain a road condition assessment result.

[0031] Specifically, based on the road condition boundary mapping features, a spatial frequency domain analysis is performed on the driving road surface to obtain a road condition evaluation result. This process is carried out through a deeper analysis by converting the image information from the spatial domain to the frequency domain. First, the system utilizes the road condition boundary mapping features obtained through the Hough circle transformation technique in the previous step, which detail the boundary and contour information of various objects on the road surface and its surrounding environment. Next, by performing a two-dimensional Fourier transform on these boundary mapping features, the image data can be converted to the frequency domain. In the frequency domain, different frequency components represent different structural information in the image: low-frequency components usually represent large smooth areas, while high-frequency components reflect local changes such as details and edges. For example, when an electric bicycle is traveling on an urban road and there is a complex road condition ahead, including potholes, manhole covers, and traffic signs of different shapes and sizes. At this time, the system will perform a spatial frequency domain analysis based on the road condition boundary mapping features obtained from the Hough circle transformation technique. Specifically, by applying a two-dimensional Fourier transform to these boundary mapping features, the system can separate each frequency component and further analyze the physical meaning corresponding to these frequency components. The low-frequency components may reveal the overall slope and flatness of the entire section of the road, while the high-frequency components can help identify the positions and severity of specific obstacles such as potholes or manhole covers. In addition, frequency domain analysis can also help detect subtle changes in the road surface, such as the development trend of cracks or wear conditions, thus providing a more accurate road condition evaluation for the rider. Based on the above analysis results, the system can generate a comprehensive and detailed road condition evaluation report, covering not only the macroscopic road conditions but also the specific obstacle information at the microscopic level. For example, the system may prompt that there is a relatively flat section of the road ahead but with a small number of potholes, and indicate the specific positions and recommended safe driving speeds. This multi-level analysis method enables the electric bicycle to achieve more intelligent and safe navigation in a complex urban environment, significantly improving the riding experience and safety. Therefore, through spatial frequency domain analysis, the intelligent road condition recognition system of the electric bicycle not only improves the accuracy of road condition perception but also provides more scientific and reasonable driving suggestions for the rider.

[0032] In a specific embodiment, the multi-frame image acquisition of the driving road surface by the binocular camera to obtain an original road condition image sequence includes: Synchronously acquire left and right images of the driving road surface through the binocular camera to obtain a left-right image pair, and perform distortion correction on the left-right image pair to obtain a corrected left-right image pair; Perform exposure compensation on the corrected left and right image pairs, adjust the image brightness to obtain left and right image pairs with balanced brightness, and perform image registration on the left and right image pairs with balanced brightness to precisely align the left and right image pairs, thereby obtaining registered left and right image pairs; wherein, the registered left and right image pairs ensure the corresponding relationship of corresponding pixel points in space. Combine the registered left and right image pairs in chronological order to form an original road condition image sequence.

[0033] Specifically, in the intelligent road condition recognition method for electric bicycles, the process of collecting multiple frames of images of the driving road surface through the binocular camera to obtain the original road condition image sequence is a complex and crucial step. First, the system uses the binocular camera to synchronously collect the left and right images, generating a pair of left and right images. Since the positions of the two lenses of the binocular camera are different, there will be a certain parallax in the captured images, which provides a basis for subsequent depth information extraction. However, due to the limitations of the optical system, these pairs of left and right images may exhibit distortion phenomena, such as barrel distortion or pincushion distortion. Therefore, after obtaining the pair of left and right images, the system will perform distortion correction on these images to eliminate the influence brought by lens distortion and ensure the authenticity and accuracy of the images. The pair of left and right images after distortion correction can more accurately reflect the actual road conditions. Next, in order to further improve the quality and consistency of the images, the system will perform exposure compensation processing on the corrected pair of left and right images. In the actual application scenario, due to the change of lighting conditions, there may be a difference in brightness between the pair of left and right images, which will affect the subsequent image registration and feature extraction effects. Therefore, by performing exposure compensation on the pair of left and right images, the image brightness can be adjusted to make the brightness of the two images tend to be consistent, thereby obtaining a pair of left and right images with balanced brightness. This process not only improves the overall quality of the images but also lays a good foundation for subsequent image processing. After obtaining the pair of left and right images with balanced brightness, the system needs to perform image registration operations on them. The purpose of image registration is to accurately align the pair of left and right images so that the corresponding pixel points have an accurate corresponding relationship in space. Specifically, the system will calculate the feature points in the pair of left and right images and perform matching and alignment according to the position relationship of these feature points. For example, in an actual application scenario, assume that an electric bicycle is traveling on an urban road, and there is a complex road condition ahead, including potholes, manhole covers, and traffic signs, etc., which are objects with different shapes and sizes. At this time, the system extracts and matches the feature points in the pair of left and right images, finds the corresponding feature points and performs precise alignment to ensure that the position relationship of each pair of corresponding pixel points in space remains consistent. This process is crucial for subsequent parallax calculation and 3D reconstruction because only when the pair of left and right images are accurately aligned can the parallax value of each pixel point be accurately calculated, and then the corresponding depth information can be deduced. After completing the image registration, the system combines the registered pair of left and right images in chronological order to form the original road condition image sequence. This process involves the processing of image data of multiple consecutive frames, aiming to capture the dynamic changes of the road conditions during the driving process. For example, during the driving process of an electric bicycle, there may be situations where pedestrians suddenly cross the road or vehicles brake urgently in the front section. By continuously collecting multiple frames of images and arranging them in chronological order, the system can obtain a complete original road condition image sequence, which not only contains the static road surface information but also can reflect the movement trajectories and speed changes of dynamic objects.The advantage of doing this is that the system can perform further analysis on these multi-frame image sequences, such as optical flow field estimation, wavelet decomposition, etc., so as to achieve more accurate road condition recognition and evaluation. Generally speaking, through the above series of steps, the system can start from the initial images collected by the binocular camera, and after a series of processes such as distortion correction, exposure compensation, and image registration, finally generate a high-quality original road condition image sequence. This process not only ensures the accuracy and consistency of the image data, but also provides a solid foundation for subsequent road condition recognition and analysis. For example, in a complex urban traffic environment, by this method, an electric bicycle can perceive and recognize the changes in the road conditions ahead in real time, make timely responses, and avoid potential safety hazards, thus significantly improving the safety and comfort of cycling. The application of this series of technologies demonstrates the great potential of modern intelligent transportation systems in enhancing traffic safety and efficiency.

[0034] In a specific embodiment, the dynamic features of the road condition are extracted from the original road condition image sequence based on a preset optical flow field estimation algorithm, and a road surface dynamic feature map is obtained, including: Perform multi-directional gradient decomposition on the original road condition image sequence based on a preset optical flow field estimation algorithm to obtain a group of image gradient vectors, and perform light intensity consistency constraint calculation on the group of image gradient vectors to obtain pixel light intensity change mapping data; wherein, the pixel light intensity change mapping data includes the light intensity change amount in the vertical direction and the light intensity change amount in the horizontal direction; Perform spatio-temporal correlation analysis on the original road condition image sequence based on the pixel light intensity change mapping data to obtain a pixel motion vector field, and perform non-linear interpolation reconstruction on the pixel motion vector field to obtain a dense optical flow field description; Perform scene motion decomposition on the dense optical flow field description through a variational optimization method to obtain a basic road surface motion component, and perform local feature clustering on the basic road surface motion component to obtain a road condition motion feature block; Extract high-order moment features from the road condition motion feature block to obtain a motion feature description sequence; Perform multi-dimensional feature fusion on the motion feature description sequence through tensor decomposition to obtain a road surface dynamic feature map, wherein the road surface dynamic feature map includes road surface deformation features and road surface vibration features.

[0035] Specifically, in the intelligent road condition recognition method for electric bicycles, the process of extracting dynamic features from the original road condition image sequence based on a preset optical flow field estimation algorithm to obtain a road surface dynamic feature map is a complex and multi-level operation. First, the system uses the preset optical flow field estimation algorithm to perform multi-directional gradient decomposition on the original road condition image sequence, generating an image gradient vector group. This process captures local change information in the image by calculating the gradient values of each pixel point in different directions. Specifically, the system calculates the gradient values in the vertical and horizontal directions. These gradient values not only reflect the color or brightness changes of each pixel point in the image but also contain potential motion information. To ensure the consistency of these gradient values, the system performs an intensity consistency constraint calculation on the image gradient vector group, thereby obtaining pixel intensity change mapping data. This mapping data includes the vertical intensity change amount and the horizontal intensity change amount, providing a basis for subsequent dynamic feature extraction. Next, based on the pixel intensity change mapping data, the system performs spatio-temporal correlation analysis on the original road condition image sequence to obtain a pixel motion vector field. In this process, the system not only considers the changes in the spatial dimension but also combines the changes in consecutive frame images in the time dimension to accurately capture the motion trajectory of each pixel point. For example, in a practical application scenario, assume that an electric bicycle is traveling on a city road, and there is a complex road condition ahead, including pedestrians suddenly crossing the road, vehicles braking suddenly, etc. At this time, through spatio-temporal correlation analysis, the system can identify the specific motion trajectories and speed changes of these dynamic objects. To further improve the accuracy and resolution of the optical flow field, the system performs non-linear interpolation reconstruction on the pixel motion vector field to generate a dense optical flow field description. This dense optical flow field description not only contains the motion information of each pixel point but also reflects the motion trend of the entire scene. Subsequently, the system decomposes the dense optical flow field description through a variational optimization method to obtain the road surface basic motion component. This process aims to separate the motion information related to the road surface from the complex optical flow field data and exclude the influence of other dynamic interference factors. For example, in the above application scenario, the system can distinguish the motion of dynamic objects such as pedestrians and vehicles and the deformation or vibration of the road itself. To better understand these motion components, the system performs local feature clustering on the road surface basic motion component to generate road condition motion feature blocks. These blocks not only reflect the macroscopic morphological changes of the road surface but also reveal the dynamic characteristics at the microscopic level, such as potholes and cracks. Next, the system performs high-order moment feature extraction on the road condition motion feature blocks to obtain a motion feature description sequence. High-order moment feature extraction is an effective mathematical tool that can characterize the complex structural information of an image region by calculating its statistical features. For example, in the above application scenario, by performing high-order moment feature extraction on the road condition motion feature blocks, the system can obtain a detailed description of the deformation and vibration of the road surface.These descriptions not only contain static road form information but also can reflect dynamic change trends, such as the minute undulations or vibration conditions on the road surface. Finally, the system performs multi-dimensional feature fusion on the sequence of motion feature descriptions through tensor decomposition to generate a road surface dynamic feature map. Tensor decomposition is a powerful data analysis method that can decompose multi-dimensional data into multiple low-dimensional subspaces, thereby effectively extracting the key features therein. In this process, the system can not only fuse feature information from different directions and scales but also comprehensively consider the changes in the time dimension to generate a comprehensive and detailed road surface dynamic feature map. This map not only includes road surface deformation features but also covers road surface vibration features, providing important data support for subsequent road condition assessment and early warning. For example, in the above application scenario, when an electric bicycle approaches a severely damaged section of the road, the road surface dynamic feature map generated by the system through the above series of steps can accurately identify the specific damage conditions of the section, such as the location, size, and depth of potholes. At the same time, the system can also detect the subtle vibration conditions of the road surface, which may indicate potential safety hazards, such as cracks or collapses about to occur on the road surface. Based on these detailed feature information, the system can provide more scientific and reasonable driving suggestions for cyclists, such as slowing down or taking a detour, thus significantly improving the safety and comfort of cycling. In short, through the above series of complex processing steps, the intelligent road condition recognition system of the electric bicycle not only improves the accuracy of road condition perception but also provides more reliable navigation support for cyclists, demonstrating the great potential of modern intelligent transportation technology in enhancing traffic safety and efficiency.

[0036] In a specific embodiment, performing multi-scale wavelet decomposition on the road surface dynamic feature map to obtain a set of road condition feature components, including: Performing orthogonal wavelet basis decomposition on the road surface dynamic feature map to obtain a multi-level decomposition coefficient matrix, and calculating the high-frequency sub-band energy of the multi-level decomposition coefficient matrix to obtain a road surface texture feature spectrum; wherein, the road surface texture feature spectrum includes a roughness feature map and a flatness feature map; Performing frequency band decomposition on the road surface texture feature spectrum through a preset wavelet packet transform technology to obtain a road surface frequency distribution feature, and performing subspace projection reconstruction on the road surface frequency distribution feature to obtain a set of road condition feature components; wherein, the set of road condition feature components includes road surface geometric features, material features, and environmental illumination features.

[0037] Specifically, in the intelligent road condition recognition method for electric bicycles, the process of performing multi-scale wavelet decomposition on the road surface dynamic feature map to obtain the road condition feature component set is a multi-level and complex operation. First, the system performs orthogonal wavelet basis decomposition on the road surface dynamic feature map to generate a multi-level decomposition coefficient matrix. This process decomposes the image data into sub-bands of different scales by applying orthogonal wavelet basis functions, and each sub-band contains information within a specific frequency range. Specifically, orthogonal wavelet basis decomposition can decompose the road surface dynamic feature map into multiple levels, each level corresponding to a different resolution and level of detail. For example, in a practical application scenario, assume that an electric bicycle is traveling on an urban road, and there is a section of complex road conditions ahead, including potholes, manhole covers, and traffic signs, etc., which are objects with different shapes and sizes. At this time, by performing orthogonal wavelet basis decomposition on the road surface dynamic feature map, the system can extract feature information at different scales, from the macroscopic overall road shape to the microscopic local details. Next, the system calculates the high-frequency sub-band energy of the multi-level decomposition coefficient matrix to obtain the road surface texture feature spectrum. The high-frequency sub-band energy calculation aims to quantify the high-frequency components in each sub-band, and these high-frequency components usually reflect the detailed information in the image, such as edges, textures, etc. Specifically, the system calculates the energy distribution of each sub-band and maps it to the corresponding feature spectrum. For example, in the above application scenario, the system can generate a roughness feature map and a flatness feature map by calculating the energy of the high-frequency sub-bands. The roughness feature map reflects the unevenness of the road surface, while the flatness feature map describes the overall smoothness of the road. These feature spectra not only provide information about the microscopic structure of the road surface but also lay the foundation for subsequent frequency band decomposition and feature reconstruction. To further refine the road surface feature analysis, the system performs frequency band decomposition on the road surface texture feature spectrum through a preset wavelet packet transform technology to obtain the road surface frequency distribution feature. Wavelet packet transform is an extended wavelet analysis method that can decompose the signal within a finer frequency band range, thereby capturing more detailed information. In this process, the system decomposes the road surface texture feature spectrum into multiple frequency bands, and each frequency band represents the feature information within a specific frequency range. For example, in the above application scenario, the system can further decompose the roughness feature map and the flatness feature map into multiple frequency bands to reveal the feature changes within different frequency ranges. This frequency band decomposition can not only improve the accuracy of feature extraction but also better adapt to the complex and changeable road condition environment. Subsequently, the system performs subspace projection reconstruction on the road surface frequency distribution feature to obtain the road condition feature component set. Subspace projection reconstruction is a mathematical method that can effectively extract key features and reduce the data dimension by projecting high-dimensional data into a low-dimensional subspace. Specifically, the system projects the frequency distribution feature after wavelet packet transform into a suitable subspace, thereby generating a road condition feature component set containing various feature information.For example, in the above application scenario, the system can generate a set of road condition feature components including road surface geometric features, material features, and environmental lighting features through subspace projection reconstruction. The road surface geometric features describe the three-dimensional shape of the road surface, such as slope, curvature, etc.; the material features reflect the properties of the road surface materials, such as asphalt, concrete, etc.; the environmental lighting features consider the visual effects under different lighting conditions to ensure that the system can accurately identify road conditions under various lighting conditions. To better understand the application scenario of this process, assume that an electric bicycle is traveling on a busy urban road, and there is a complex road condition ahead, including a severely damaged section, some newly paved asphalt areas, and several traffic signs. In this case, the system first uses orthogonal wavelet basis decomposition to decompose the road surface dynamic feature map into a multi-level decomposition coefficient matrix, and extracts the feature information at different scales. Then, by calculating the high-frequency sub-band energy of these decomposition coefficient matrices, the system generates a roughness feature map and a flatness feature map, which detail the microscopic structure and overall smoothness of the road surface. For example, the system can identify the specific location and severity of the damaged section and evaluate its impact on riding safety. Then, the system performs band decomposition on the road surface texture feature spectrum through wavelet packet transform technology to further refine the feature analysis. This step enables the system to accurately capture the feature changes in different frequency ranges, thereby providing more detailed road condition information. For example, the system can detect the differences between the newly paved asphalt areas and the old road surface and identify potential safety hazards. Finally, through subspace projection reconstruction of the road surface frequency distribution features, the system generates a set of road condition feature components including road surface geometric features, material features, and environmental lighting features. These feature components not only provide information about the road surface shape and materials but also consider the visual effects under different lighting conditions to ensure that the system can accurately identify road conditions in various environments. In summary, through the above series of complex processing steps, the intelligent road condition recognition system of the electric bicycle not only improves the accuracy of road condition perception but also provides more reliable navigation support for riders. For example, in the above application scenario, when the electric bicycle approaches a severely damaged section, the set of road condition feature components generated by a series of operations such as orthogonal wavelet basis decomposition, high-frequency sub-band energy calculation, wavelet packet transform, and subspace projection reconstruction can accurately identify the specific damage conditions of the section, such as the location, size, and depth of the potholes. At the same time, the system can also detect the subtle vibration conditions of the road surface, which may indicate potential safety hazards, such as cracks or collapses about to occur on the road surface. Based on these detailed feature information, the system can provide more scientific and reasonable driving suggestions for riders, such as slowing down or taking a detour, thereby significantly improving the safety and comfort of riding. In short, through the above series of complex processing steps, modern intelligent transportation technologies have demonstrated great potential in enhancing traffic safety and efficiency.

[0038] In a specific embodiment, the regional division of the road condition feature component set by the adaptive threshold segmentation technology to obtain the road condition type feature region includes: Perform multi-dimensional feature decomposition on the road condition feature component set to obtain a road surface feature scale tensor, and perform local neighborhood analysis on the road surface feature scale tensor to obtain a road surface area gradient mapping matrix; Perform spatial correlation calculation on the road surface area gradient mapping matrix through a preset autocorrelation function to obtain a regional correlation feature map, and perform local variance estimation on the regional correlation feature map to obtain a regional heterogeneity description matrix; Through the adaptive threshold segmentation technology, perform adaptive threshold calculation on the regional correlation feature map based on the regional heterogeneity description matrix to obtain a multi-level threshold sequence, and perform region growing segmentation on the regional heterogeneity description matrix based on the multi-level threshold sequence to obtain an initial segmentation region set; Perform regional morphological processing on the initial segmentation region set to obtain a regional contour feature sequence, and perform boundary optimization reconstruction based on the regional contour feature sequence to obtain a reconstructed region boundary map, and perform regional topological relationship analysis on the reconstructed region boundary map to obtain a regional spatial association map; Perform regional label mapping on the regional spatial association map through a preset regional label mapping algorithm to obtain a road condition type feature region; wherein, the road condition type feature region includes a determined flat region, a concave-convex region, and an obstacle region.

[0039] Specifically, in the intelligent road condition recognition method for electric bicycles, the process of dividing the road condition feature component set through adaptive threshold segmentation technology to obtain the road condition type feature region is complex and precise. First, the system performs multi-dimensional feature decomposition on the road condition feature component set to obtain a road surface feature scale tensor, and performs local neighborhood analysis on the road surface feature scale tensor to generate a road surface region gradient mapping matrix. This process involves decomposing complex road condition information into multiple dimensions to more carefully analyze the feature changes in each dimension. For example, in an application scenario, assume an electric bicycle is traveling on an urban road, and there is a section of road ahead that contains different types of road conditions, such as a smooth asphalt section, potholes, and obstacles. In this case, by performing multi-dimensional feature decomposition on the road condition feature component set, the system can extract the feature information in each dimension, including the geometric shape of the road surface, material properties, and environmental lighting conditions, and then generate a road surface region gradient mapping matrix to describe the gradient changes between different regions. Next, the system calculates the spatial correlation of the road surface region gradient mapping matrix through a preset autocorrelation function to obtain a region correlation feature map, and further estimates the local variance of the region correlation feature map to obtain a region heterogeneity description matrix. In this process, the autocorrelation function is used to evaluate the similarity and difference between adjacent regions to reveal the internal connection and change law between different regions. For example, in the above application scenario, the system can generate a region correlation feature map by calculating the spatial correlation of the road surface region gradient mapping matrix, which can show the mutual relationship between different regions, such as which regions have similar texture or slope characteristics. Then, the system obtains a region heterogeneity description matrix by estimating the local variance of the region correlation feature map. This step aims to quantify the degree of change of the eigenvalue within each region, thereby helping the system better understand the feature distribution within different regions. Subsequently, the system uses adaptive threshold segmentation technology to perform adaptive threshold calculation on the region correlation feature map based on the region heterogeneity description matrix to obtain a multi-level threshold sequence, and performs region growing segmentation on the region heterogeneity description matrix based on the multi-level threshold sequence to obtain an initial segmentation region set. The key in this stage is to determine an appropriate threshold to distinguish different road condition types. For example, in the above application scenario, the system can automatically adjust the threshold according to the data in the region heterogeneity description matrix so that those regions with obvious differences (such as a flat section and a pothole area) can be accurately distinguished. This adaptive method not only improves the accuracy of segmentation but also enhances the robustness of the system, enabling it to work effectively in various environments.Next, the system performs regional morphological processing on the set of initial segmentation regions to obtain a sequence of regional contour features, and based on the sequence of regional contour features, performs boundary optimization and reconstruction to obtain a reconstructed regional boundary map, and analyzes the regional topological relationship of the reconstructed regional boundary map to obtain a regional spatial association map. In this process, regional morphological processing is used to improve the quality of the segmentation result and ensure that the boundaries of each region are clear and continuous. For example, in the above application scenario, the system can eliminate noise and irregular boundaries in the set of initial segmentation regions through regional morphological processing, so as to obtain a more accurate sequence of regional contour features. Then, the system performs boundary optimization and reconstruction based on these feature sequences to generate a reconstructed regional boundary map, and further analyzes the topological relationship between regions to obtain a regional spatial association map. This step is crucial for understanding the mutual position relationship between different regions, and it helps the system to comprehensively grasp the structure and layout of the entire road condition. Finally, the system performs regional label mapping on the regional spatial association map through a preset regional label mapping algorithm to obtain a road condition type feature region; wherein, the road condition type feature region includes a determined flat region, a concave-convex region, and an obstacle region. The role of the regional label mapping algorithm is to assign corresponding labels to different regions according to all the information obtained previously, for the convenience of subsequent processing and decision-making. For example, in the above application scenario, the system can classify each region as a flat region, a concave-convex region, or an obstacle region according to the data in the regional spatial association map and in combination with the preset criteria. This step not only enables the system to clearly identify the specific road condition type, but also provides it with basic data support, enabling the rider to make more scientific and reasonable driving decisions based on this information. For example, when an electric bicycle approaches a severely damaged section of the road, the system identifies the specific damage situation of the section through the above series of operations, and provides detailed road condition information, such as the location, size, and depth of the potholes, and can also detect the subtle vibration of the road surface, which may indicate potential safety hazards. Based on this information, the system can suggest that the rider slow down or choose another route, thus improving the safety and comfort of riding. In short, through this series of complex processing steps, modern intelligent transportation technology not only improves the accuracy of road condition perception, but also provides more reliable navigation support for riders.

[0040] In a specific embodiment, the boundary extraction and contour fitting of the road condition type feature region based on the Hough circle transformation technology to obtain the road condition boundary mapping feature includes: Performing polar coordinate transformation on the road condition type feature region through a preset Hough circle transformation technology to obtain a polar coordinate feature matrix, and performing circular arc feature detection on the polar coordinate feature matrix to obtain a set of road surface curvature features; wherein, the set of road surface curvature features includes road surface circular arc curvature and road surface boundary curvature; Edge refinement is performed on the set of road surface curvature features through non-maximum suppression to obtain a refined edge feature map, and gradient direction statistics are performed on the refined edge feature map to obtain an edge direction histogram; Based on the edge direction histogram, circular contour voting is performed on the refined edge feature map to obtain a circular contour candidate set, and boundary point extraction is performed on the circular contour candidate set to obtain a boundary feature point sequence; Ellipse fitting transformation is performed on the boundary feature point sequence to obtain a contour fitting parameter set, and boundary curve reconstruction is performed based on the contour fitting parameter set to obtain a road condition boundary curve group; Shape feature extraction is performed on the road condition boundary curve group through a preset geometric invariant moment technique to obtain a boundary shape descriptor, and feature clustering analysis is performed on the boundary shape descriptor to obtain a road condition boundary mapping feature; wherein, the road condition boundary mapping feature includes a road surface contour mapping feature and a road surface boundary mapping feature.

[0041] Specifically, in the intelligent road condition recognition method, the process of extracting the boundary and fitting the contour of the road condition type feature area based on the Hough circle transform technology to obtain the road condition boundary mapping feature is delicate and complex. First, the system performs a polar coordinate transformation on the road condition type feature area through the preset Hough circle transform technology to obtain a polar coordinate feature matrix, and further performs an arc feature detection on the polar coordinate feature matrix to generate a road surface curvature feature set. Among them, the road surface curvature feature set includes information such as the road surface arc curvature and the road surface boundary curvature. For example, in a specific scenario, assume an electric bicycle is traveling on a curve with different radii of curvature. In this case, the system uses the Hough circle transform technology to convert the road condition type feature area from the Cartesian coordinate system to the polar coordinate system to better analyze circular or approximately circular road features. By performing an arc feature detection on the polar coordinate feature matrix, the system can identify different curvature changes in the road, such as sharp turns and gentle curves, which helps the rider understand the road conditions ahead in advance and make corresponding driving decisions. Subsequently, the system uses the non-maximum suppression technology to refine the edges of the road surface curvature feature set to generate a refined edge feature map, and further performs a gradient direction statistics on the refined edge feature map to obtain an edge direction histogram. This step aims to improve the accuracy of edge detection and ensure that every possible circular or elliptical shape is accurately captured. For example, in the above application scenario, when the system detects different curvatures of the curve, the non-maximum suppression technology can remove some unnecessary edge information, making the remaining edges clearer and more accurate. Then, the system performs a gradient direction statistics on these refined edges to form an edge direction histogram, which shows the edge distribution in different directions and provides basic data support for the subsequent circular contour voting. Next, based on the edge direction histogram, a circular contour voting is performed on the refined edge feature map to generate a circular contour candidate set, and boundary points are extracted from the circular contour candidate set to obtain a boundary feature point sequence. The key in this stage is to use the circular contour voting algorithm to determine the pixel points most likely to represent the circular contour according to the information in the edge direction histogram. For example, in the above application scenario, the system can screen out the boundary points that form circular or elliptical road features through the voting mechanism according to the data provided by the edge direction histogram, and then construct multiple circular contour candidate sets. This process not only improves the accuracy of circular contour detection but also lays a foundation for subsequent boundary extraction and contour fitting. After that, the system performs an ellipse fitting transformation on the boundary feature point sequence to obtain a contour fitting parameter set, and based on the contour fitting parameter set, a boundary curve reconstruction is performed to obtain a road condition boundary curve group. In this process, the ellipse fitting transformation is used to optimize the positions of the boundary feature points to make them more conform to the real road boundary shape.For example, in the above application scenario, through elliptical fitting transformation, the system can adjust the positions of the boundary feature points, making the finally generated road condition boundary curve group closer to the actual road edge shape. In this way, not only can more accurate road condition information be provided, but it can also help cyclists better understand the specific layout and direction of the road ahead. Finally, the system extracts the shape features of the road condition boundary curve group through the preset geometric invariant moment technology to obtain the boundary shape descriptors, and performs feature clustering analysis on the boundary shape descriptors to obtain the road condition boundary mapping features; among them, the road condition boundary mapping features include the road surface contour mapping feature and the road surface boundary mapping feature. The geometric invariant moment technology is used here to calculate the feature vectors of the boundary shape, and these vectors are invariant to rotation, scaling, and translation, so they are very suitable for describing shape features. For example, in the above application scenario, the system can extract the shape features of the road condition boundary curve group through the geometric invariant moment technology to form a series of boundary shape descriptors. Then, through feature clustering analysis of these descriptors, the system can identify different types of road condition boundary features, such as the edges of flat sections, the boundaries of pothole areas, and the contours around obstacles. These detailed road condition boundary mapping features provide important reference information for cyclists, enabling them to drive more safely in complex and changeable road environments. In short, through the above series of processing steps, modern intelligent transportation technology not only improves the accuracy of road condition perception but also greatly improves the driving experience of cyclists.

[0042] In a specific embodiment, the spatial frequency domain analysis of the driving road surface based on the road condition boundary mapping features to obtain a road condition evaluation result includes: Perform bilateral spectrum decomposition on the road condition boundary mapping features to obtain a frequency feature tensor, and calculate the band energy of the frequency feature tensor to obtain a road surface vibration spectrum distribution map; wherein, the road surface vibration spectrum distribution map includes a vertical vibration frequency spectrum and a horizontal vibration frequency spectrum; Extract local features from the road surface vibration spectrum distribution map through joint time-frequency analysis to obtain a time-frequency feature sequence, and accumulate the sub-band energy of the time-frequency feature sequence to obtain a road condition frequency feature matrix; Based on the road condition frequency feature matrix, perform spectral peak detection on the time-frequency feature sequence to obtain a set of characteristic frequencies, and perform frequency clustering analysis on the set of characteristic frequencies to obtain a road condition spectrum pattern map; wherein, the road condition spectrum pattern map includes a steady frequency pattern and a mutation frequency pattern; Perform spatial phase correlation analysis on the road condition spectrum pattern map to obtain a set of phase correlation features, and perform spatial frequency reconstruction based on the set of phase correlation features to obtain a road surface frequency response map; Through the spatial spectral density estimation algorithm, perform road condition evaluation on the driving road surface based on the road surface frequency response map to obtain a road condition evaluation result.

[0043] Specifically, in the intelligent road condition recognition method for electric bicycles, the process of performing spatial frequency domain analysis on the driving road surface based on the road condition boundary mapping features to obtain the road condition evaluation result is a multi-level and complex operation. First, the system performs bilateral spectral decomposition on the road condition boundary mapping features to generate a frequency feature tensor, and further calculates the band energy of the frequency feature tensor to obtain the road surface vibration spectrum distribution map. Among them, the road surface vibration spectrum distribution map includes the vertical vibration spectrum and the horizontal vibration spectrum. This process aims to transform the road condition boundary mapping features from the spatial domain to the frequency domain to better analyze their frequency characteristics. For example, in a specific scenario, assume an electric bicycle is traveling on a road with different vibration characteristics, such as slight undulations or potholes in some sections. In this case, by performing bilateral spectral decomposition on the road condition boundary mapping features, the system can extract the vibration information in different frequency bands to form a frequency feature tensor. Then, by calculating the band energy of these frequency feature tensors, the system can generate the road surface vibration spectrum distribution map, which shows the vibration characteristics of the road surface in the vertical and horizontal directions. Next, the system performs local feature extraction on the road surface vibration spectrum distribution map through joint time-frequency analysis to generate a time-frequency feature sequence, and further accumulates the sub-band energy of the time-frequency feature sequence to obtain the road condition frequency feature matrix. The key in this stage is to use time-frequency analysis technology to capture the dynamic characteristics of the road surface vibration changing with time and frequency. For example, in the above application scenario, when the electric bicycle passes through a road section with a complex vibration pattern, the system can extract the local vibration characteristics at each time point through joint time-frequency analysis technology and organize them into a time-frequency feature sequence. Then, by accumulating the sub-band energy of these time-frequency feature sequences, the system can generate the road condition frequency feature matrix, which contains the energy distribution in different frequency bands and provides data support for subsequent feature detection and clustering analysis. Subsequently, the system performs spectral peak detection on the time-frequency feature sequence based on the road condition frequency feature matrix to generate a set of characteristic frequencies, and further performs frequency clustering analysis on the set of characteristic frequencies to obtain the road condition spectrum pattern map. Among them, the road condition spectrum pattern map includes a steady frequency pattern and a mutation frequency pattern. The goal of this stage is to identify different types of frequency patterns to better understand the changing rules of the road conditions. For example, in the above application scenario, when the electric bicycle passes through a relatively flat road section, the system can identify the steady frequency pattern through spectral peak detection technology, indicating that the road surface is relatively smooth. When the vehicle passes through potholes or obstacles, the system can detect the mutation frequency pattern, which reflects the irregular changes of the road surface. Through frequency clustering analysis, the system can classify these different frequency patterns and generate the road condition spectrum pattern map to provide detailed road condition information for the rider.Next, the system performs a spatial phase correlation analysis on the road condition spectrum pattern diagram to generate a phase correlation feature group, and based on the phase correlation feature group, performs a spatial frequency reconstruction to obtain a road surface frequency response map. This process aims to further optimize the expression of road condition information by analyzing the phase relationships between different frequency patterns. For example, in the above application scenario, when an electric bicycle is traveling on a road with complex vibration patterns, the system can identify the phase correlation feature group between different frequency patterns through spatial phase correlation analysis. Then, based on these phase correlation feature groups, the system can perform a spatial frequency reconstruction to generate a more accurate road surface frequency response map. This map not only reflects the overall vibration characteristics of the road surface but also reveals the subtle changes in local areas, providing a more comprehensive road condition perception for the rider. Finally, the system uses a spatial spectral density estimation algorithm to perform a road condition assessment on the driving road surface based on the road surface frequency response map, generating a road condition assessment result. The spatial spectral density estimation algorithm is used here to quantify the frequency distribution characteristics in the road surface frequency response map, thereby generating the final road condition assessment result. For example, in the above application scenario, when the electric bicycle is approaching a severely damaged section of the road, the system can analyze the frequency distribution in the road surface frequency response map through the spatial spectral density estimation algorithm to generate a detailed road condition assessment result. These assessment results not only include the overall flatness of the road surface but also can reveal potential safety hazards, such as the location, size, and depth of potholes, etc. Based on this information, the system can provide scientific and reasonable driving suggestions for the rider, such as slowing down or choosing another route, thus significantly improving the safety and comfort of riding. In summary, through the above series of complex processing steps, modern intelligent transportation technologies not only improve the accuracy of road condition perception but also provide more reliable navigation support for riders. For example, in the above application scenario, when an electric bicycle is traveling on a road with different curvature radii and vibration characteristics, the system generates a detailed road condition assessment result through a series of operations such as bilateral spectrum decomposition, joint time-frequency analysis, spectral peak detection, spatial phase correlation analysis, and spatial spectral density estimation. These results not only help the rider understand the specific conditions of the road ahead in advance but also provide important reference information, enabling them to drive more safely in a complex and changing road environment. In short, through this series of delicate operations, modern intelligent transportation technologies demonstrate great potential and greatly improve the riding experience and safety of riders.

[0044] The intelligent road condition recognition method for an electric bicycle in the embodiment of the present invention has been described above. Next, the intelligent road condition recognition device for an electric bicycle in the embodiment of the present invention will be described. Please refer to Figure 2 , an embodiment of the intelligent road condition recognition device for an electric bicycle in the embodiment of the present invention includes: The acquisition module 21 is configured to collect multiple frames of images of the driving road surface through the binocular camera to obtain an original road condition image sequence; The extraction module 22 is configured to perform dynamic feature extraction on the original road condition image sequence based on a preset optical flow field estimation algorithm to obtain a road surface dynamic feature map; The decomposition module 23 is configured to perform multi-scale wavelet decomposition on the road surface dynamic feature map to obtain a set of road condition feature components; The partitioning module 24 is configured to perform region partitioning on the set of road condition feature components through an adaptive threshold segmentation algorithm to obtain road condition type feature regions; The fitting module 25 is configured to perform boundary extraction and contour fitting on the road condition type feature regions based on the Hough circle transform technique to obtain road condition boundary mapping features; The analysis module 26 is configured to perform spatial frequency domain analysis on the driving road surface based on the road condition boundary mapping features to obtain a road condition evaluation result.

[0045] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to that described in the above method embodiment, and details are not described herein again.

[0046] Refer to Figure 3 , and in an embodiment of the present invention, a computer device is further provided. The internal structure of the computer device may be as Figure 3 shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected through a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program, when executed by the processor, implements the above method.

[0047] Those skilled in the art can understand that Figure 3 the structure shown in

[0048] is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0049] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium provided by the present invention and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0050] It should be noted that in this article, the terms "including", "comprising", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the statement "including one..." does not exclude the presence of additional identical elements in the process, apparatus, article, or method that includes the element.

[0051] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structural or equivalent process transformation made by using the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. An intelligent road condition recognition method for an electric bicycle, characterized in that, A binocular camera is provided on the electric bicycle, including the following steps: Collect multiple frames of images of the driving road surface through the binocular camera to obtain an original image sequence of the road conditions; Extract dynamic features from the original image sequence of the road conditions based on a preset optical flow field estimation algorithm to obtain a road surface dynamic feature map; Perform multi-scale wavelet decomposition on the road surface dynamic feature map to obtain a set of road condition feature components; Perform regional division on the set of road condition feature components through an adaptive threshold segmentation algorithm to obtain a road condition type feature region; Extract the boundary and contour fitting of the road condition type feature region based on the Hough circle transformation technology to obtain a road condition boundary mapping feature; Perform spatial frequency domain analysis on the driving road surface based on the road condition boundary mapping feature to obtain a road condition evaluation result.

2. The intelligent road condition recognition method for an electric bicycle according to claim 1, wherein, The step of collecting multiple frames of images of the driving road surface through the binocular camera to obtain an original image sequence of the road conditions includes: Synchronously collect left and right images of the driving road surface through the binocular camera to obtain a pair of left and right images, and perform distortion correction on the pair of left and right images to obtain a pair of corrected left and right images; Perform exposure compensation on the pair of corrected left and right images to adjust the image brightness to obtain a pair of left and right images with balanced brightness, and perform image registration on the pair of left and right images with balanced brightness to accurately align the pair of left and right images to obtain a pair of registered left and right images; wherein, the pair of registered left and right images ensures the corresponding relationship of corresponding pixel points in space; Combine the pair of registered left and right images in chronological order to form an original image sequence of the road conditions.

3. The intelligent road condition recognition method of the electric bicycle according to claim 1, characterized in that, The step of extracting dynamic features from the original image sequence of the road conditions based on a preset optical flow field estimation algorithm to obtain a road surface dynamic feature map includes: Perform multi-directional gradient decomposition on the original image sequence of the road conditions based on a preset optical flow field estimation algorithm to obtain a group of image gradient vectors, and perform light intensity consistency constraint calculation on the group of image gradient vectors to obtain pixel light intensity change mapping data; wherein, the pixel light intensity change mapping data includes the vertical direction light intensity change amount and the horizontal direction light intensity change amount; Perform spatio-temporal correlation analysis on the original image sequence of the road conditions based on the pixel light intensity change mapping data to obtain a pixel motion vector field, and perform non-linear interpolation reconstruction on the pixel motion vector field to obtain a dense optical flow field description; Perform scene motion decomposition on the dense optical flow field description through a variational optimization method to obtain a basic road surface motion component, and perform local feature clustering on the basic road surface motion component to obtain a road condition motion feature block; Extract high-order moment features from the road condition motion feature block to obtain a motion feature description sequence; Perform multi-dimensional feature fusion on the motion feature description sequence through tensor decomposition to obtain a road surface dynamic feature map, wherein the road surface dynamic feature map includes road surface deformation features and road surface vibration features.

4. The intelligent road condition recognition method for an electric bicycle according to claim 1, wherein, The step of performing multi-scale wavelet decomposition on the road surface dynamic feature map to obtain a set of road condition feature components includes: Perform orthogonal wavelet basis decomposition on the road surface dynamic feature map to obtain a multi-level decomposition coefficient matrix, and calculate the high-frequency sub-band energy of the multi-level decomposition coefficient matrix to obtain a road surface texture feature spectrum; wherein, the road surface texture feature spectrum includes a roughness feature map and a flatness feature map. Perform frequency band decomposition on the road surface texture feature spectrum through a preset wavelet packet transform technology to obtain the road surface frequency distribution feature, and perform subspace projection reconstruction on the road surface frequency distribution feature to obtain a road condition feature component set; wherein, the road condition feature component set includes road surface geometric features, material features, and environmental lighting features.

5. The intelligent road condition recognition method for an electric bicycle according to claim 1, wherein Perform regional division on the road condition feature component set through an adaptive threshold segmentation technology to obtain road condition type feature regions, including: Perform multi-dimensional feature decomposition on the road condition feature component set to obtain a road surface feature scale tensor, and perform local neighborhood analysis on the road surface feature scale tensor to obtain a road surface area gradient mapping matrix. Perform spatial correlation calculation on the road surface area gradient mapping matrix through a preset autocorrelation function to obtain a regional correlation feature map, and perform local variance estimation on the regional correlation feature map to obtain a regional heterogeneity description matrix. Through an adaptive threshold segmentation technology, perform adaptive threshold calculation on the regional correlation feature map based on the regional heterogeneity description matrix to obtain a multi-level threshold sequence, and perform region growing segmentation on the regional heterogeneity description matrix based on the multi-level threshold sequence to obtain an initial segmentation region set. Perform regional morphological processing on the initial segmentation region set to obtain a regional contour feature sequence, and perform boundary optimization reconstruction based on the regional contour feature sequence to obtain a reconstructed region boundary map, and perform regional topological relationship analysis on the reconstructed region boundary map to obtain a regional spatial association map. Perform regional label mapping on the regional spatial association map through a preset regional label mapping algorithm to obtain road condition type feature regions; wherein, the road condition type feature regions include determined flat regions, concave and convex regions, and obstacle regions.

6. The intelligent road condition recognition method for an electric bicycle according to claim 1, characterized in that, Perform boundary extraction and contour fitting on the road condition type feature regions based on the Hough circle transform technology to obtain road condition boundary mapping features, including: Perform polar coordinate transformation on the road condition type feature regions through a preset Hough circle transform technology to obtain a polar coordinate feature matrix, and perform circular arc feature detection on the polar coordinate feature matrix to obtain a road surface curvature feature set; wherein, the road surface curvature feature set includes road surface circular arc curvature and road surface boundary curvature. Perform edge refinement on the road surface curvature feature set through non-maximum suppression to obtain a refined edge feature map, and perform gradient direction statistics on the refined edge feature map to obtain an edge direction histogram. Perform circular contour voting on the refined edge feature map based on the edge direction histogram to obtain a circular contour candidate set, and perform boundary point extraction on the circular contour candidate set to obtain a boundary feature point sequence. Perform elliptical fitting transformation on the sequence of boundary feature points to obtain a set of contour fitting parameters, and reconstruct the boundary curve based on the set of contour fitting parameters to obtain a group of road condition boundary curves; Extract the shape features of the group of road condition boundary curves through the preset geometric invariant moment technology to obtain a boundary shape descriptor, and perform feature clustering analysis on the boundary shape descriptor to obtain the road condition boundary mapping features; wherein, the road condition boundary mapping features include road surface contour mapping features and road surface boundary mapping features.

7. The intelligent road condition recognition method for an electric bicycle according to claim 1, characterized in that, Perform spatial frequency domain analysis on the driving road surface based on the road condition boundary mapping features to obtain a road condition evaluation result, including: Perform bilateral spectrum decomposition on the road condition boundary mapping features to obtain a frequency feature tensor, and calculate the band energy of the frequency feature tensor to obtain a road surface vibration spectrum distribution map; wherein, the road surface vibration spectrum distribution map includes a vertical vibration frequency spectrum and a horizontal vibration frequency spectrum; Extract local features of the road surface vibration spectrum distribution map through joint time-frequency analysis to obtain a time-frequency feature sequence, and accumulate the sub-band energy of the time-frequency feature sequence to obtain a road condition frequency feature matrix; Perform spectral peak detection on the time-frequency feature sequence based on the road condition frequency feature matrix to obtain a set of characteristic frequencies, and perform frequency clustering analysis on the set of characteristic frequencies to obtain a road condition frequency spectrum pattern map; wherein, the road condition frequency spectrum pattern map includes a stable frequency pattern and a mutation frequency pattern; Perform spatial phase correlation analysis on the road condition frequency spectrum pattern map to obtain a set of phase correlation features, and perform spatial frequency reconstruction based on the set of phase correlation features to obtain a road surface frequency response map; Perform road condition evaluation on the driving road surface based on the road surface frequency response map through a spatial spectral density estimation algorithm to obtain a road condition evaluation result.

8. An intelligent road condition recognition device for an electric bicycle, characterized in that, A binocular camera is provided on the electric bicycle, including: An acquisition module for acquiring a sequence of original road condition images of the driving road surface through the binocular camera; An extraction module for extracting dynamic features of the sequence of original road condition images based on a preset optical flow field estimation algorithm to obtain a road surface dynamic feature map; A decomposition module for performing multi-scale wavelet decomposition on the road surface dynamic feature map to obtain a set of road condition feature components; A partitioning module for partitioning the set of road condition feature components through an adaptive threshold segmentation algorithm to obtain a road condition type feature region; A fitting module for extracting the boundary and fitting the contour of the road condition type feature region based on the Hough circle transformation technology to obtain the road condition boundary mapping features; An analysis module for performing spatial frequency domain analysis on the driving road surface based on the road condition boundary mapping features to obtain a road condition evaluation result.

9. A computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images