Non-motor vehicle overspeed identification method and device based on image sequence

Abstract

The invention provides a non-motor vehicle overspeed identification method and device based on an image sequence, and relates to the technical field of image analysis, and the method comprises the steps: obtaining image sequence data collected by a non-motor vehicle in a low-illumination curve channel; feature parameter extraction is carried out based on the image sequence data, mapping processing is carried out based on the extracted feature parameters, and an embedded track point set of the non-motor vehicle in the curve channel is obtained; performing direction sensitive tensor graph construction and multi-scale steering perception graph modeling processing to obtain a multi-dimensional graph structure; a speed estimation vector sequence is obtained through processing of an angular graph attention mechanism and a convolution trajectory transformation network; carrying out acceleration filtering and compensation processing to obtain a speed sequence without turning inertia errors; and performing curve overspeed behavior analysis to obtain a judgment result of whether the non-motor vehicle has the curve overspeed behavior or not. According to the method, fine reconstruction of the non-motor vehicle track, inertial error filtering and speed behavior analysis are realized by utilizing multi-stage algorithm combination.

Description

Technical Field

[0001] This invention relates to the field of image analysis technology, and more specifically, to a method and apparatus for identifying non-motorized vehicles speeding based on image sequences. Background Technology

[0002] Currently, non-motorized vehicles (such as electric bicycles and shared bicycles) are widely used on urban roads and have become an important part of urban transportation. With their increasing usage frequency, the traffic hazards caused by speeding in complex traffic environments are becoming increasingly prominent, especially in low-light underground passages, enclosed multi-level tunnels, and narrow spaces with curves, where speeding can easily lead to serious traffic accidents such as collisions and slips. Existing non-motorized vehicle speed detection methods mostly rely on single-frame image target detection, manually set speed thresholds, or electronic speed limit devices. These methods have two significant problems: first, it is difficult to accurately estimate the true speed trajectory of non-motorized vehicles in continuous image sequences, especially when turning, obstructed, or in blurred images; second, without preset speed limits or road markings, it is impossible to effectively distinguish between normal riding and risky speeding behavior. Furthermore, traditional motion modeling methods cannot accurately represent the curvature changes, turning trends, and inertial interference of non-motorized vehicles in curved road sections, resulting in low accuracy in speed anomaly identification, with false alarms and missed alarms, making it difficult to meet the actual needs of refined traffic behavior management in complex scenarios.

[0003] Therefore, there is an urgent need for a method and device for identifying non-motorized vehicles speeding based on image sequences to solve the above-mentioned technical problems. Summary of the Invention

[0004] The purpose of this invention is to provide a method and apparatus for identifying non-motorized vehicle speeding based on image sequences, so as to improve the above-mentioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows: Firstly, this application provides a method for identifying non-motorized vehicle speeding based on image sequences, including: Acquire image sequence data of non-motorized vehicles in low-light curved passages; Based on the image sequence data, non-motorized vehicle boundary features and channel structure parameters are extracted, and based on the extracted features and channel structure parameters, mapping processing is performed to obtain the set of embedded trajectory points of non-motorized vehicles in the curved channel; The embedded trajectory point set is processed by constructing a direction-sensitive tensor graph and modeling a multi-scale steering perception graph to obtain a multi-dimensional graph structure containing information on time dimension, spatial direction and steering curvature. The multidimensional graph structure is processed by an angular graph attention mechanism and a convolutional trajectory transformation network to obtain a speed estimation vector sequence for non-motorized vehicles; The velocity estimation vector sequence is subjected to acceleration filtering and compensation processing to obtain a velocity sequence with the turning inertia error removed; The speed sequence is analyzed to determine whether the non-motorized vehicle is speeding on curves.

[0005] Secondly, this application also provides a non-motorized vehicle speeding recognition device based on image sequences, comprising: The acquisition unit is used to acquire image sequence data of non-motorized vehicles in low-light curved passages; The mapping unit is used to extract non-motorized vehicle boundary features and channel structure parameters based on the image sequence data, and to perform mapping processing based on the extracted features and channel structure parameters to obtain the set of embedded trajectory points of non-motorized vehicles in the curved channel. The modeling unit is used to construct a direction-sensitive tensor graph and perform multi-scale steering perception graph modeling processing on the embedded trajectory point set to obtain a multi-dimensional graph structure containing time dimension, spatial direction and steering curvature information. The processing unit is used to process the multidimensional graph structure through an angular graph attention mechanism and a convolutional trajectory transformation network to obtain a speed estimation vector sequence for non-motorized vehicles; The removal unit is used to perform acceleration filtering and compensation processing on the velocity estimation vector sequence to obtain a velocity sequence with the turning inertia error removed. The analysis unit is used to analyze the speed sequence for non-motorized vehicles to determine whether they exhibit speeding behavior on curves.

[0006] The beneficial effects of this invention are as follows: This invention addresses scenarios such as low-light conditions, enclosed underground passages, and irregular roads with continuous curves. It proposes a complete technical process that integrates orientation-sensitive graph structure modeling, curvature-guided manifold mapping, angular attention mechanism, trajectory convolution estimation, and speed anomaly discrimination fusion algorithm. By acquiring image sequence data and constructing a multi-dimensional graph model with spatial orientation, temporal continuity, and structural curvature information, it utilizes a multi-stage algorithm combination to achieve fine reconstruction of non-motorized vehicle trajectories, inertial error filtering, and speed behavior analysis. Ultimately, without the need for preset speed thresholds or manual restrictions, it achieves adaptive recognition of speeding behavior on curved road sections, significantly improving the system's robustness, accuracy, and applicability in complex environments. This overcomes the core defects of existing technologies, such as rule dependence, one-sided recognition, and poor adaptability.

[0007] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0008] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic diagram of the non-motorized vehicle speeding identification method based on image sequences as described in this embodiment of the invention; Figure 2 This is a schematic diagram of the non-motorized vehicle speeding recognition device based on image sequence as described in an embodiment of the present invention.

[0010] In the diagram: 701, Acquisition Unit; 702, Mapping Unit; 703, Modeling Unit; 704, Processing Unit; 705, Removal Unit; 706, Analysis Unit. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0012] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0013] Example 1: This embodiment provides a method for identifying non-motorized vehicles speeding based on image sequences.

[0014] See Figure 1 The figure shows that the method includes steps S1, S2, S3, S4, S5 and S6.

[0015] Step S1: Acquire image sequence data of non-motorized vehicles in low-light curved passages; It is understood that the image sequence data in this step includes not only image frames under consecutive timestamps, but also non-motorized vehicle edge feature extraction signals under low light conditions, structural modeling parameters of the passage space (such as curve curvature, passage width, inclination, and wall material), and light reflection models in the environment (such as the reflection coefficients of the ground and walls, and the reflection trajectory affected by the lighting arrangement). The image frames under consecutive timestamps come from image acquisition devices (such as low-light cameras, infrared imaging devices, or wide dynamic range cameras) installed inside the enclosed underground passage. The non-motorized vehicle edge feature extraction signals under low light conditions refer to the signals obtained by introducing low-light image enhancement algorithms (such as low-light Retinex enhancement, global contrast enhancement, and infrared contour compensation) during the image acquisition stage. The geometric modeling parameters and light reflection models of the passage structure are usually obtained in advance through construction design drawings or on-site scanning modeling systems (such as laser SLAM or 3D point cloud modeling), and then modeling is performed using these data. This section provides a raw data foundation that combines temporal continuity, structural clarity, and spatial positioning availability for subsequent boundary feature extraction and trajectory embedding modeling. It also constructs a perception input port that is compatible with low light and complex terrain, thereby significantly improving the system's adaptability and stability in non-ideal visual environments.

[0016] Step S2: Extract the boundary features and channel structure parameters of non-motorized vehicles based on the image sequence data, and perform mapping processing based on the extracted features and channel structure parameters to obtain the set of embedded trajectory points of non-motorized vehicles in the curved channel; It is understandable that this step extracts core features with geometric and motion significance from the enhanced image frames and structural parameters provided during the image acquisition stage to establish an embeddable motion trajectory relationship between non-motorized vehicles and the curved passage. This step not only transforms the blurred and directionally unstable edge information in the image frames into a spatially continuous and structurally reliable set of motion trajectory points, but also introduces physical modeling of the passage structure throughout the process. This gives the trajectory point set a measurable geometric basis in subsequent velocity modeling, solving existing problems such as large motion path reconstruction errors, severe boundary distortion, and non-physical pixel trajectories in complex curved structures. In this step, step S2 includes steps S21, S22, S23, and S24.

[0017] Step S21: Perform boundary tensor response extraction processing on the image sequence data. By constructing a multi-frequency guided edge tensor response model of the non-motorized vehicle candidate region in each image frame of the image sequence data, a preliminary non-motorized vehicle boundary response map is obtained. Understandably, this step introduces a multi-frequency guided edge tensor response model, decomposing the image signal in different frequency domains. In the low-frequency region, it suppresses reflective areas and shadow interference, while in the high-frequency region, it enhances the boundary contour response, especially the structural change features along the motion direction. Specifically, this step uses wavelet transform or Gabor filtering to construct a multi-frequency tensor map of the image. Based on this, it combines motion candidate regions (regions potentially containing non-motorized vehicles pre-screened through inter-frame differencing or optical flow estimation) to guide the edge response, thus constructing a boundary response tensor with both direction sensitivity and frequency domain stability.

[0018] This processing not only targets the static pixel intensity in image frames but also incorporates the temporal evolution characteristics of the image sequence. A temporal consistency check mechanism is used to further filter boundary results, eliminating false edges caused by instantaneous bright spots or abrupt changes in background structure, while preserving the structural contour evolution trend of non-motorized vehicles in consecutive frames. The final preliminary non-motorized vehicle boundary response map not only contains spatial edge distribution information but also includes directional features and confidence labels, providing a high-quality structural foundation for subsequent boundary curvature modeling and channel geometry mapping.

[0019] Step S22: Perform local boundary curvature estimation processing on the preliminary non-motorized vehicle boundary response map. By performing multi-scale direction fitting on the non-motorized vehicle boundary points and introducing a second-order guidance term in the strong turning region to suppress image noise interference, a set of boundary curvature points is obtained. Understandably, this step first performs multi-scale orientation fitting on the set of boundary points detected in the boundary response map. Within the neighborhood of each candidate boundary point, an orientation fitting window is constructed, and curvature approximation is performed at multiple scales (e.g., 3, 5, 7, and 11 pixel neighborhoods) using polynomial curve fitting or local Bézier curve models. The first-order directional derivative vector is extracted, and initial curvature is estimated by combining it with the local boundary orientation. The multi-scale strategy is particularly crucial in this scenario: because in low-light curves, non-motorized vehicles may experience localized stillness (leading to overly smooth edges) or rapid turning (leading to edge deformation) within a single frame. Different scales can address the conflicting requirements of noise suppression and boundary preservation.

[0020] To further enhance the geometric stability of the fitting results, this step introduces a second-order guidance term in the strong turning region (i.e., the boundary segment with drastic curvature changes and directional fluctuations). This is achieved by constructing a curvature acceleration term between boundary points to correct the directional drift of the first-order fitting and suppress local anomalous jumps in the image caused by reflections, shadows, or texture interference. Essentially, this guidance mechanism adds a regularization term to the curvature estimation function, enabling it to follow the boundary without causing oscillations in the turning segment. This results in a set of boundary curvature points with smooth direction, realistic curvature, and continuous boundaries, which are then used as the boundary features for non-motorized vehicles.

[0021] Step S23: Perform channel structure-guided perspective geometry modeling on the boundary curvature point set. By using the channel structure parameters of the image sequence data to perform multi-stage mapping from the pixel domain to the curvature space on the trajectory in the image frame, the estimated motion path sequence of the non-motorized vehicle is obtained. Understandably, this step utilizes camera perspective parameters (including focal length, tilt angle, and projection center) and channel structural parameters (such as curve radius, channel width, and wall orientation) from the image sequence data. The channel structural parameters are derived from uploaded channel construction data collected by staff. An initial affine mapping matrix is ​​constructed from image pixels to the world coordinate system, and a basic spatial projection is performed on the boundary curvature point set. This mapping is used to roughly eliminate projection distortion, ensuring the spatial distribution of boundary points matches the actual channel structure. Subsequently, for channel regions with strong curvature, a nonlinear curvature inversion model is introduced: the curvature information of each boundary point in the image frame is fitted with the "geometric curve profile function" in the channel parameters to construct a mapping relationship from image coordinates to curvature space, giving each trajectory point a realistic spatial interpretation of local curvature trends.

[0022] It is understandable that the geometric curve profile function in this step is as follows: ; in, Indicates the path in the curve length parameter The geometric curvature value at that point, This represents the arc length parameter along the non-motorized vehicle's path. Indicates the path is in The spatial curvature at that point.

[0023] Furthermore, this step introduces temporal consistency constraints, tracking the spatial transformation trend of boundary curvature points in the image sequence. By fusing the movement patterns of the same point's trajectory across multiple frames, a dynamic alignment function for trajectory points across different frames is constructed, thereby enabling the motion trajectory to form a continuous and derivable path shape in physical space. These mapping outputs ultimately form a set of predicted motion path sequences arranged in temporal order, possessing spatial consistency and physical structural basis.

[0024] Step S24: Perform manifold embedding processing on the motion path sequence to obtain the set of embedded trajectory points of non-motorized vehicles in the curve channel.

[0025] Understandably, this step first constructs a similarity graph structure for trajectory points based on the temporal adjacency relationships, local curvature change rates, and spatial distribution density between path points. The local neighborhood of each point not only considers temporal continuity but also introduces spatial turning trends as weight adjustment factors to characterize the embedding differences between different motion modes such as "accelerated turning" and "gentle curve gliding" within the path. Based on this graph structure, this step employs the Riemannian manifold embedding algorithm to project the nonlinearly distributed path points in the high-dimensional space into the embedding space. This enhances the ability to express local motion structures (such as sharp curves and velocity jumps) while maintaining the global path topology.

[0026] Meanwhile, to avoid local overfitting or graph breakage during the embedding process, this step introduces an equidistant mapping mechanism to ensure that the physical distribution consistency between path points is maintained in the embedding space. That is, high-density segments, such as at the entrance of a bend, remain concentrated, rather than being artificially stretched after embedding, thereby improving the interpretability of the embedded path and the stability of subsequent graph construction.

[0027] Step S3: Construct a direction-sensitive tensor graph and model a multi-scale steering perception graph from the embedded trajectory point set to obtain a multi-dimensional graph structure containing time dimension, spatial direction and steering curvature information; It is understandable that this step, based on the trajectory point set obtained after manifold embedding, further constructs a multidimensional tensor graph structure that integrates time series, spatial orientation, and geometric curvature information. This graph structure is not only used to carry the spatial connection relationship of the trajectory itself, but also serves as the core data support form for subsequent speed estimation and behavior recognition. This step significantly improves the model's adaptability to behavioral pattern variability, especially in effectively distinguishing between behaviors such as low-speed large curves and fast small curves. In this step, step S3 includes steps S31, S32, S33, and S34.

[0028] Step S31: Perform local direction encoding on the embedded trajectory point set, wherein the motion direction between adjacent trajectory points is represented as a directed edge in a tensor to obtain an initial direction tensor graph. Understandably, this system first traverses each pair of temporally adjacent trajectory points in the embedded trajectory point set, calculates their spatial position difference vector using the Euclidean distance method, extracts the unit direction vector (e.g., constructing direction angles based on changes in three-dimensional space), and then encodes this direction information as edge weights or edge attributes into the trajectory graph structure. This ensures that each edge in the graph not only connects two trajectory points but also explicitly carries information about "from which point to which direction of movement." This process essentially establishes a structural representation layer with a direction field overlaid on the trajectory graph, where each edge can be viewed as a directional motion element path.

[0029] Unlike traditional undirected trajectory graphs or simple time-series point sets, orientation-encoded tensor graphs can capture important behavioral features such as local orientation abrupt changes, continuous turning trends, and motion inertia trends during subsequent processing, making them particularly suitable for identifying complex dynamic patterns in turning regions. Furthermore, the system introduces a tensor structure (i.e., multi-channel encoding) while constructing the graph. Each tensor channel can represent different dimensions of orientation correlation, such as the principal direction, normal vector, and historical direction mean, further enhancing the graph structure's ability to represent local behavioral features.

[0030] Step S32: Perform spatial orientation consistency optimization processing on the initial orientation tensor graph, wherein the orientation average field in the spatial neighborhood is introduced into the initial orientation tensor graph, and it is smoothed and irregularly turned to change the region to obtain the optimized tensor graph structure. Understandably, this step first defines a spatial neighborhood window with a fixed radius in the embedding space for each trajectory point, extracts all neighboring trajectory points within this window that are connected to the point by graph edges, and collects their direction vectors. Then, it calculates the weighted average direction of the neighborhood direction vectors, where the weights can be adjusted based on the time difference of the trajectory points (more weight is given to points with closer time), curvature similarity (penalizing points with abrupt changes in direction), or local density (avoiding misleading information from sparse regions), forming a direction estimation field that represents the "expected motion trend" of the point.

[0031] Next, this step uses the average direction field to adjust the edge attributes in the initial tensor graph that exhibit anomalous directional abrupt changes. For edges whose directions deviate excessively from the neighborhood mean, the system replaces their direction vectors with the mean direction from the direction field, or uses regularized interpolation for smooth adjustment. Ultimately, the directional characteristics of the connections between all trajectory points are significantly optimized while maintaining the temporal order, resulting in an optimized tensor graph structure with continuous edge attributes, high directional consistency, and strong robustness to local perturbations.

[0032] Step S33: Perform multi-scale steering response modeling on the optimized tensor graph structure. By performing tensor domain hierarchical modeling on the steering gradient information at different scales, extract the feature response corresponding to the preset steering mode to obtain the steering perception graph structure. Understandably, this step first defines multiple orientation-aware scales (e.g., 3 nodes for short scales, 5-7 nodes for medium scales, and 11 or more nodes for long scales) on the optimized tensor graph structure. At each scale, this step extracts the orientation change sequence (i.e., orientation gradient) of consecutive point pairs in the trajectory, calculates indices such as the rate of change of turning angle, cumulative angle increment, and local orientation change density, and constructs an orientation gradient tensor layer for that scale. Each scale layer is essentially an orientation response map with a specific field of view, and its edge attributes describe the turning trend and intensity of change of the trajectory within that scale range.

[0033] Subsequently, this step performs hierarchical modeling of these multi-scale layers in the tensor domain: by constructing a cross-scale response stacking structure, the feature responses of each scale are combined according to scale weights and structural stability to form a multi-scale turning response map with hierarchical perception capabilities. In order to extract behavioral features with recognition significance, this step performs pattern matching on each scale layer. Based on the previously defined typical turning pattern templates (such as sharp turns, continuous fine-tuning, compound turns, etc.), structural similarity matching and response extraction are performed. Finally, the identified turning structure patterns are integrated back into the corresponding trajectory segments in the tensor map to generate a turning perception map structure with structural labels and multi-scale directional response features.

[0034] Step S34: Perform multi-dimensional graph structure fusion processing on the steering perception graph structure. By fusing the trajectory timestamp, position coordinates, direction vector and local curvature encoding into high-order graph node features, a multi-dimensional graph structure containing time dimension, spatial direction and steering curvature information is obtained.

[0035] Understandably, this step first expands the feature dimensions of each trajectory node (i.e., an embedded trajectory point) in the steering perception map. Specifically, the following four types of information are introduced into each node and jointly encoded: Timestamp feature: Each trajectory point corresponds to a frame time in the image sequence, representing its relative temporal position throughout the trajectory evolution process. This feature is used to maintain the temporal order of trajectory points and is the temporal basis for recognizing dynamic behaviors such as velocity changes and acceleration trends.

[0036] Position coordinate features: The coordinate values ​​of the trajectory points in the embedding space, preserving their geometric positions constructed in the low-dimensional curvature space, supporting spatial consistency analysis and path geometric alignment.

[0037] Direction vector features: These represent the direction of movement of non-motorized vehicles at that point. They originate from the edge attributes of the direction tensor graph in the preceding steps and are the core information for analyzing behavioral inertia, turning trends, and momentum continuity.

[0038] Local curvature coding: obtained from the boundary curvature estimation stage, used to quantify the curvature of the path at that point, especially for identifying spatial regions with specific dynamic risks such as "sharp turns" and "gliding segments".

[0039] The four types of information mentioned above are integrated into a multi-dimensional vector through a high-order node feature concatenation mechanism. Based on this, the system constructs a multi-channel tensor representation of the graph nodes, meaning that each node simultaneously contains its temporal position, directionality, geometric shape, and spatial state. To maintain dimensional consistency and representational stability among the features, this step also performs normalization and scaling on each channel (e.g., standardizing time and normalizing the direction vector), enabling features of different dimensions to collaboratively participate in learning within the graph neural network without exhibiting a dominant dimension bias problem.

[0040] Ultimately, the high-dimensional feature tensors of all nodes constitute the node attribute set of the entire multidimensional graph structure, while the edge attributes inherit the aforementioned information such as direction weights and turning amplitudes, forming a graph structure with four-fold coupling of space, time, direction, and curvature.

[0041] Step S4: Process the multidimensional graph structure through the angular graph attention mechanism and convolutional trajectory transformation network to obtain the speed estimation vector sequence of non-motorized vehicles; It is understandable that this step, based on the trajectory graph structure constructed in the previous step, which contains multi-dimensional information such as time, spatial direction, and curvature, uses a combined angular graph attention mechanism and a convolutional trajectory transformation network to perform deep modeling of the graph. To derive the speed estimation vector sequence of non-motorized vehicles from complex trajectory behaviors, this step utilizes the structure perception capability of graph neural networks and the direction selection capability of the angular attention mechanism. For the diverse, nonlinear, and frequently disturbed trajectory behaviors in curved passages, this step introduces a direction-aware graph attention mechanism and a curvature-adaptive convolutional trajectory network. This allows the system to extract high-precision, low-fluctuation speed estimation results from the graph structure, exhibiting strong robustness to abnormal structural changes. Especially in curved passages, it avoids speed misestimation or false acceleration, providing a highly reliable speed foundation for subsequent acceleration filtering and abnormal behavior recognition. In this step, step S4 includes steps S41, S42, S43, and S44.

[0042] Step S41: Initialize the attention weights of the multidimensional graph structure by angular partial derivative guidance. The initial weight matrix of graph attention is obtained by using the angular offset of each node in the multidimensional graph structure in its directional neighborhood as the kernel weight value for attention calculation. Understandably, for any trajectory node in the graph, this step first extracts the direction vector of each neighboring node from its directional neighborhood (i.e., the set of nodes pointed to by the directed edges connecting to that node), and calculates the angle between this vector and the current direction vector of the trajectory node to form an angular offset. The system then constructs an attention kernel function based on an angular partial derivative function (such as a cosine function or a Gaussian direction kernel) to map the angular offset to the corresponding attention intensity. In this step, nodes with similar angles have higher directional consistency and should be given higher weight during information fusion; while neighboring nodes with large directional abrupt changes may correspond to turning jitter, slippage, or discontinuous trajectory segments, and their attention weights will be weakened or eliminated.

[0043] To further improve the stability of directional weight calculation, this step also introduces a normalization mechanism for directional weights (such as softmax or L1 normalization) to ensure that the sum of the adjacency weights of each node is 1, forming a learnable attention distribution. This initial weight matrix serves as the basic input to the graph attention mechanism and will be used as the basis for dynamic adjustment of edge weights in the subsequent graph neural network propagation and velocity estimation process.

[0044] Step S42: Perform node neighborhood feature aggregation processing on the initial weight matrix of the graph attention. By fusing direction, time and curvature information in the variable neighborhood of the node, an aggregated feature tensor is formed to obtain a set of node state vectors. Understandably, this step performs weighted aggregation on each trajectory node in the graph, utilizing the feature data from its neighboring node set. Each neighboring node's features contain three core channels: a direction vector, reflecting the current trajectory's movement trend; a timestamp, describing the trajectory point's temporal evolution; and a local curvature value, measuring the path's curvature at that node. The system weights these features by using attention coefficients associated with each neighboring node in the guiding weight matrix, performing the following weighted combination: ; in, For adjacent nodes The nonlinear projection operation on the original feature vectors includes modules such as feature scaling and channel normalization, used to unify the representation scales of different physical quantities (direction as a vector, time as a scalar, curvature as a scalar). Through this operation, this step generates a direction-time-structure coupled state vector at each node. It accurately reflects the geometric state, evolution trend, and local path complexity of the current trajectory point in its motion context. Represents a node For adjacent nodes Attention coefficient Represents a node The set of adjacent nodes, Indicates adjacent nodes The original feature vector, This represents the coupled node state vector.

[0045] In this step, the aggregation process adopts a variable neighborhood mechanism, that is, the size of the neighborhood of a node is not fixed, but dynamically adjusted according to the trajectory density, directional continuity and structural consistency in the graph. This allows for more detailed aggregation to be built in dense path segments (to adapt to high-frequency directional changes), while expanding the aggregation field of view in sparse regions to enhance stability.

[0046] Step S43: Perform trajectory convolution transformation on the set of node state vectors. By performing sliding window convolution operation based on direction sequence on the sequential time axis of the trajectory, and introducing channel curvature as a convolution weight adjustment factor, a convolution feature map reflecting the dynamics of the trajectory time series is obtained. Understandably, this step first reorders the state vectors of all nodes in the trajectory graph according to their timestamps, constructing a time-ordered sequence of state vectors. Then, a fixed-size sliding window (e.g., length 5) is defined, and a one-dimensional convolution operation is performed by sliding along the time axis on this sequence. Unlike standard one-dimensional convolution, the convolution kernel in this system not only slides in the time dimension but also incorporates the trend of direction vector changes within each window for directional consistency weighting. That is, if the direction changes drastically within a window, the direct contribution of that segment of the convolution output to the velocity estimation is reduced to avoid misleading results caused by sharp turns or path disturbances.

[0047] Furthermore, this step introduces a channel curvature adjustment factor as an external control parameter for the convolution weights. Specifically, for each sliding window, the average curvature value of its nodes is calculated as an indicator of the geometric complexity of that region. The system dynamically adjusts the convolution kernel weight response based on the curvature magnitude—enhancing the perception of directional gradient changes in high-curvature (i.e., sharp bends) regions and enhancing the filtering effect on velocity stability in low-curvature regions. This "structure-guided convolution" mechanism can be viewed as a geometric modulation method for the trajectory convolution process, allowing the feature map to retain path structure differences during convolution.

[0048] The final output convolutional feature map is a time-ordered tensor structure containing the comprehensive feature response of each trajectory point in its local time-direction neighborhood. This feature map not only encodes the motion trend of the non-motorized vehicle at each moment, but also reflects its historical evolution path and the upcoming change of direction, providing a context-aware dynamic representation for subsequent speed estimation.

[0049] Step S44: Perform velocity vector analysis on the convolutional feature map. By performing first and second derivative analysis on the node sequence in the convolutional output domain, obtain the velocity estimation vector sequence of the non-motorized vehicle.

[0050] Understandably, this step first performs first-order derivative analysis on the node state sequence in the convolutional feature map to capture the velocity change trend of the trajectory between consecutive time points. Since the feature vector of each node contains direction, position, and temporal semantics, the system uses the timestamp as the main variable of the derivative to calculate the rate of change of each node's features between adjacent time points, obtaining its instantaneous velocity estimate. This process is equivalent to establishing a velocity trend curve in the feature space, whose slope corresponds to the average velocity of the object on that path segment, while incorporating the nonlinear component brought about by directional changes, so that the estimation not only depends on the position difference but also reflects the influence of trajectory shape changes on the velocity.

[0051] Secondly, this step further performs second-order derivative analysis on the velocity sequence, that is, differentiating the rate of change of velocity over time to obtain the acceleration trend, which is used to identify inertial-driven, rapid acceleration, or deceleration behaviors. This higher-order information is particularly important in curved sections because non-motorized vehicles often exhibit short-term "gliding" or "delayed speed drop" phenomena during turns, and first-order estimation alone may misjudge the true speed state. Through second-order derivative analysis, this step can identify and suppress such speed fluctuations caused by inertial drive, improving the physical accuracy and behavioral consistency of speed estimation.

[0052] In this step, each velocity value in the entire velocity vector sequence has a clear temporal position, structural source, and directional coupling background, forming a set of high-confidence, behavior-aware velocity estimates.

[0053] Step S5: Perform acceleration filtering and compensation processing on the velocity estimation vector sequence to obtain a velocity sequence with the turning inertia error removed; It is understandable that this step, through curvature-sensing acceleration filtering and trajectory structure-driven speed compensation, effectively solves the problem of severe interference from directional fluctuations and inertial slippage in speed estimation in curved environments, improves the steady-state, reliability and structural consistency of the speed sequence, provides high-quality and error-controllable input signals for the behavior recognition module, and significantly enhances the reliability of the system in actual traffic scenarios. In this step, step S5 includes steps S51, S52, S53 and S54.

[0054] Step S51: Perform local acceleration response extraction processing on the velocity estimation vector sequence. Calculate the acceleration at each time node in the velocity estimation vector sequence and use the direction switching rate as a suppression factor to obtain the original acceleration vector sequence of the non-motorized vehicle. Understandably, in this step, the system calculates the rate of change of velocity for each pair of adjacent velocity estimation points to obtain the preliminary instantaneous acceleration. This process not only considers numerical changes but also simultaneously extracts the trajectory direction vectors corresponding to these two time points and calculates the rate of change of their included angle, defined as the direction switching rate. This is one of the key innovations of this step: during turning, U-turns, or coasting, although the numerical velocity changes significantly, these changes are often caused by direction switching or inertia, rather than actual acceleration behavior.

[0055] Therefore, this step introduces the direction switching rate as a dynamic suppression factor into the acceleration analysis: if the acceleration is high in a certain period, but the direction switching rate is also high, it indicates that the acceleration may be generated by trajectory redirection rather than dynamic drive, and its reliability should be reduced; conversely, in regions where the direction remains stable but the velocity suddenly increases, the acceleration is retained as a valid motion feature. This suppression mechanism is implemented through a threshold suppression function, which essentially uses directional stability as a weighting function for acceleration reliability to filter out misjudgments caused by directional jumps.

[0056] Ultimately, the original acceleration vector sequence output in this step contains both the acceleration response at all time points and the directional coupling background of each acceleration, enabling subsequent filtering and compensation operations to make judgments based on structural information.

[0057] Step S52: Perform inertial disturbance identification processing on the original acceleration vector sequence. By setting a steering inertial detection interval in the high curvature segment of the embedded trajectory, identify the non-real speed fluctuation area dominated by the turning trajectory structure and obtain the inertial disturbance time period mark set. Understandably, this step first extracts all high-curvature segments from the embedded trajectory structure, i.e., trajectory regions with curvature values ​​exceeding a set threshold within continuous trajectory points. These high-curvature regions typically correspond to sharp turns, continuous bends, or rapid rotations, and are spatial areas prone to inertial disturbances. Subsequently, the system sets a steering inertial detection interval within each high-curvature segment. The interval length can be dynamically adjusted based on the local time span of the trajectory, the rate of curvature change, and the intensity of direction switching. This interval not only includes the peak curvature position but also extends before and after it for a certain time window to cover the entire propagation stage of inertial influence. Inertial disturbances include the lateral centripetal acceleration signal generated by the vehicle's turning motion. This is an interference signal that contaminates the accelerometer's true reading in the direction of the device's movement, leading to errors in speed calculation. The purpose of this step is to mark these "unreliable" data segments, providing a basis for subsequent data processing.

[0058] Within these detection intervals, this step combines the original acceleration vector sequence to analyze the amplitude, trend, and directional consistency of acceleration anomalies, identifying "pseudo-fluctuation segments" that do not conform to the dynamic driving law. For example, when the acceleration significantly deviates from the neighborhood average and the direction of change is consistent with the path curvature orientation, but there is no continuous velocity growth trend, this segment can be marked as an inertial disturbance. Finally, a set of inertial disturbance period markers is generated, containing all suspected non-true velocity anomaly time periods caused by steering structures, and each marker is accompanied by its corresponding curvature interval and disturbance intensity score.

[0059] This step utilizes the geometric curvature distribution of the trajectory to perform structural constraint analysis on velocity fluctuations. The system can accurately identify the "non-dynamic velocity disturbance" region caused by the superposition of nonlinear path morphology and motion inertia, thereby improving the invention's ability to distinguish between "real velocity behavior" and "path-induced errors".

[0060] Step S53: Perform velocity smoothing sequence reconstruction processing on the inertial disturbance time period mark set. By weighting the velocity values ​​within the marked time period, a preliminary velocity smoothing sequence of non-motorized vehicles is obtained. Understandably, this step first defines the reconstruction range for each marked disturbance period, i.e., the velocity subsequence including the start and end boundaries of that period. Then, it performs a weighted estimation of the velocity values ​​within the subsequence based on three dimensions: Directional stability weighting: Evaluate the rate of change of direction corresponding to each velocity point within the disturbance segment. The more stable the direction, the higher the weight. Regions with abrupt changes in direction are considered to be strongly affected by inertial disturbances and should be given lower weights to reduce their impact on the velocity fitting curve.

[0061] Neighborhood velocity trend consistency weighted estimation: By comparing the velocity trends with those of the preceding and following unperturbed regions, it is determined whether the perturbed segment deviates from the overall motion trend. Points consistent with the trend are given higher confidence, while points deviating from the trend are given lower weights, thereby achieving continuity restoration of trajectory velocity.

[0062] Local curvature adjustment factor weighted estimation: Regions with high curvature are more susceptible to disturbances. This step uses a pre-set penalty factor based on curvature values ​​to automatically reduce the velocity response of high curvature points and enhance the dynamic adaptation to changes in path morphology.

[0063] The weights of the above three dimensions are preset according to the corresponding thresholds. Then, this step uses a weighted sliding window averaging method or a first-order fitting interpolation method to re-estimate the velocity values ​​within the perturbation segment. During the reconstruction process, it is ensured that the velocity curve maintains a consistent slope and trend at the beginning and end of the perturbation segment and the undisturbed region, avoiding new discontinuities or abrupt velocity changes in the reconstruction results. For some coasting segments or areas with significant direction changes, Bézier smoothing or spline curve interpolation can also be used to enhance the naturalness of the transition.

[0064] Ultimately, the preliminary velocity smoothing sequence output in this step not only preserves the true velocity change trend in the trajectory, but also eliminates fluctuation anomalies caused by path morphology or inertial errors, providing a high-quality, low-noise velocity foundation for the next step of global velocity curve recalibration and behavior determination.

[0065] Step S54: Perform global trend calibration processing on the preliminary velocity smoothing sequence, wherein, under the premise of ensuring energy conservation, the overall velocity curve is refitted to obtain a velocity sequence with the turning inertia error removed.

[0066] Understandably, the initial speed smoothing sequence is first subjected to a global trend analysis to assess the speed change trend across the entire trajectory range. In particular, for areas with large-scale changes such as cornering, deceleration, and acceleration, this step introduces a physical constraint to ensure that the speed change is consistent with the physical energy conservation.

[0067] Next, this step performs a global fitting of the entire velocity sequence, generating a smooth velocity curve that conforms to the laws of physical motion using a fitting algorithm (such as least squares, Bézier curve fitting, or spline interpolation). During fitting, this step dynamically adjusts the fitting weights based on the velocity changes in the turning areas. Especially in areas with high curvature or sharp turns, the system assigns higher weights to the fitting model to ensure smoother velocity changes in these areas.

[0068] This step also incorporates local correction factors during the fitting process to specifically correct for inertial errors that may occur during turning or deceleration. This ensures that speed changes during turning are reasonably corrected without affecting the speed representation on straight sections or smooth driving sections. Finally, a speed sequence with the inertia error of turning is obtained. This sequence can truly reflect the actual speed of non-motorized vehicles on curves and other complex road sections. It removes short-term speed fluctuations caused by inertial gliding or path turning, ensuring that the speed estimate is smoother and more reliable.

[0069] Step S6: Analyze the speed sequence for non-motorized vehicles to determine whether they exhibit speeding behavior on curves.

[0070] It is understandable that this step, through comprehensive analysis of speed sequences and road structure, can accurately determine whether non-motorized vehicles are speeding on curves and can adapt to complex road environments in real time. Compared to traditional speeding detection methods, this step not only considers fixed speed thresholds but also combines road geometry, the actual dynamic behavior of non-motorized vehicles, and temporal consistency analysis, thereby improving the accuracy and sensitivity of speeding detection. Especially in complex curves and low-light conditions, it can effectively identify speeding behavior and avoid false positives. This provides a more precise analytical tool for traffic monitoring, traffic safety management, and intelligent transportation systems. In this step, step S6 includes steps S61, S62, S63, and S64.

[0071] Step S61: Encode the velocity sequence by jointly encoding each velocity value in the velocity sequence with the spatial curvature and directional change rate of its corresponding embedded trajectory point into an embedding vector to obtain the velocity representation set of non-motorized vehicles; Understandably, this step extracts the velocity value and the corresponding spatial curvature (i.e., the path's curvature at that point) for each trajectory point. Curvature reflects the non-linearity of the path. For example, the curvature is greater in curves, while it approaches zero on straight sections. By jointly encoding the velocity value and the curvature value, the velocity representation of each node can not only reflect the instantaneous speed of motion but also describe the geometric characteristics of that point on the path. Nodes with high curvature values ​​(such as sharp turns) receive more attention related to their velocity because the motion of such nodes is typically more risky.

[0072] In addition to spatial curvature, this step also uses the rate of change of direction at the node (i.e., the rate of change of steering angle) as an important dimension of the encoding. The rate of change of direction measures the degree of steering aggression of a non-motorized vehicle within a specific time period. For example, the rate of change of direction is higher during sharp turns, while it is close to zero during smooth driving.

[0073] For each trajectory point, this step calculates the relationship between the rate of change of direction and the time interval, and encodes it as one of the features of that point along with the speed. This feature helps the system understand whether the non-motorized vehicle's motion is smooth, and whether there are behaviors such as acceleration, deceleration, or coasting. Especially in the turning area of ​​a curve, the rate of change of direction can indicate possible inertial effects or steering force during the motion.

[0074] Finally, by jointly encoding the velocity value, curvature, and rate of change of direction, this information is integrated into a high-dimensional embedding vector. In this way, each trajectory point not only contains its instantaneous velocity information, but also incorporates its spatial and turning features, forming an embedding representation rich in multidimensional information.

[0075] Step S62: Perform category distribution learning processing on the speed representation set. Estimate the normal curve speed embedding distribution in the preset historical trajectory and calculate the degree of deviation of the current speed representation set under the distribution to obtain a density deviation score vector that reflects the degree of speed anomaly. Understandably, this step first utilizes the normal speed sequences from historical trajectory data under similar curve environments to construct a speed embedding distribution model. This model includes speed values ​​and their corresponding embedding vectors under different curve types and speed ranges. Through statistical analysis of the characteristics of these normal trajectories, this step can learn the typical motion patterns of non-motorized vehicles in various curve scenarios and construct a speed embedding space distribution model, which is a probability density function (such as Gaussian distribution or multidimensional kernel density estimation). In this process, the invention will group historical trajectories according to different curve types, vehicle behavior patterns (such as fast turns, smooth driving, etc.), and environmental factors (such as road friction, speed limits, etc.) and learn the speed distribution characteristics of each group. Once the current velocity representation set is obtained, this step compares these velocity embedding vectors with the normal velocity embedding distribution in the historical trajectory. Specifically, for each current velocity vector, its probability density value under the normal velocity distribution is calculated, i.e., the degree of matching between the current velocity and the historical distribution. If the current velocity is near the center of the distribution, it indicates normal behavior; if the current velocity deviates significantly from the center of the normal distribution, it suggests that the velocity value may indicate abnormal behavior.

[0076] The formula for calculating its probability density value under the normal velocity distribution is as follows: ; in, This represents the probability density value of the current velocity vector under the normal velocity distribution; Represents the current velocity vector. The dimension of the velocity vector. It is pi. The determinant of the covariance matrix of the velocity vectors represents the variance and correlation between features; The inverse of the covariance matrix is ​​used to weight the deviations of the velocity vector, reflecting the relative importance of different features. This is the mean vector of the normal velocity distribution, representing the average level of various velocity characteristics in the historical trajectory. It is the transpose symbol. This represents the square of the Mahalanobis distance, which measures the distance between the current speed and the normal speed distribution. The larger the value of this term, the greater the difference between the current speed and historical normal behavior, and the smaller the probability density value.

[0077] This step calculates a density deviation score based on these matching scores. The density deviation score is the standardized inverse ratio of the probability density, and it represents the degree of deviation of the current velocity representation set from the historical normal velocity distribution. A lower deviation score indicates that the current velocity is close to the historical normal trajectory, while a higher deviation indicates an abnormal velocity.

[0078] By calculating the offset of all nodes, this step yields a density deviation score vector, which reflects the degree of velocity anomaly at each time point. Nodes with higher scores typically indicate velocity anomalies, possibly due to speeding, inertial coasting, or miscalculations caused by path structure.

[0079] Step S63: Perform behavioral state classification processing on the density deviation score vector. By classifying the high offset value regions in the density deviation score vector, a set of local behavioral state labels for a preset time period is obtained. Understandably, this step sets a scoring threshold based on the aforementioned density deviation score to distinguish between normal and abnormal behavior. For example, when the density deviation score exceeds a certain preset threshold, it indicates that there may be abnormal speed behavior during that period, such as excessive acceleration or deviation from the normal trajectory. The system categorizes the offset value into multiple levels, such as: low offset value (score close to 0): indicating that the current speed is within the normal range and there is no abnormal behavior; medium offset value: indicating some abnormal fluctuations, possibly caused by inertial gliding or other short-term disturbances; high offset value: indicating significant speed changes, possibly indicating speeding, rapid acceleration, or other sudden behaviors.

[0080] Based on the classification results, this step generates local behavior status labels for each time period. For example: in the low offset range (normal state), the label is "normal driving"; in the medium offset range (suspicious behavior), the label is "abnormal fluctuation" or "potential coasting"; in the high offset range (abnormal behavior), the label is "speeding" or "abnormal acceleration".

[0081] Finally, this step matches the local behavioral state labels for each time period with the corresponding trajectory points and time periods based on the behavioral classification results. Each behavioral label not only represents a trajectory segment but also associates it with the specific time period and behavioral intensity of that segment, enabling subsequent analysis to be precise down to the time period level and providing rich information for subsequent decision-making.

[0082] Step S64: Perform global trajectory determination processing on the set of local behavior state labels. By determining whether each local state label in the set of local behavior state labels is connected or continuous in the trajectory graph structure, the determination result of whether the non-motorized vehicle has speeding behavior on curves is obtained.

[0083] Understandably, this step first evaluates the temporal continuity of each label in the local behavior state label set. For example, if the "speeding" label appears multiple times in a trajectory and these labels are consecutive in the time series, it indicates that the behavior is likely a continuous speeding behavior, rather than a single speed anomaly. This step also calculates the persistence of the local behavior labels within a time window, that is, whether the state labels exhibit similar patterns in multiple adjacent time periods. For example, on a curve, if the "speeding" label persists for multiple time frames, the behavior is judged to be continuous speeding; if the "speeding" label appears only occasionally, it may only be a local anomaly.

[0084] Next, this step associates the local behavior state labels with nodes in the trajectory graph structure. Through connectivity analysis, it can be determined whether different local behavior labels form a continuous chain of behaviors in the trajectory graph. For example, if a speeding behavior label and a subsequent speed anomaly label are connected in the trajectory graph and exhibit temporal continuity, then these labels may reflect a complete speeding behavior trajectory. Specifically, this involves checking whether the various local behavior states in the label set can be connected by edges (representing the connection relationship between trajectory points) or continue continuously in time within the graph structure, using path connectivity in the graph to determine whether there is a continuous speeding behavior.

[0085] Finally, based on connectivity and persistence analysis, this step combines the criteria for determining cornering speeding to comprehensively assess whether cornering speeding behavior exists. If, on a certain corner segment of the trajectory, there are prolonged and continuous "speeding" behavior labels, and these labels form a continuous chain of behavior through connected paths in the trajectory map, the system will determine that the trajectory contains cornering speeding behavior. This step will also assess whether the speeding behavior meets the speeding standard based on its duration and intensity. If the speeding behavior occurs instantaneously on the corner, it may only be an isolated incident, but if it is a continuous high-deviation behavior, it is judged as risky speeding behavior.

[0086] Example 2: like Figure 2 As shown, this embodiment provides a non-motorized vehicle speeding recognition device based on image sequences. See [link to relevant documentation]. Figure 2 The device includes an acquisition unit 701, a mapping unit 702, a modeling unit 703, a processing unit 704, a removal unit 705, and an analysis unit 706.

[0087] The acquisition unit 701 is used to acquire image sequence data of non-motorized vehicles in low-light curved passages; The mapping unit 702 is used to extract non-motorized vehicle boundary features and channel structure parameters based on the image sequence data, and to perform mapping processing based on the extracted features and channel structure parameters to obtain the set of embedded trajectory points of non-motorized vehicles in the curved channel; Modeling unit 703 is used to construct a direction-sensitive tensor graph and perform multi-scale steering perception graph modeling processing on the embedded trajectory point set to obtain a multi-dimensional graph structure containing time dimension, spatial direction and steering curvature information. Processing unit 704 is used to process the multidimensional graph structure through an angular graph attention mechanism and a convolutional trajectory transformation network to obtain a speed estimation vector sequence for non-motorized vehicles; The removal unit 705 is used to perform acceleration filtering and compensation processing on the velocity estimation vector sequence to obtain a velocity sequence with the turning inertia error removed. The analysis unit 706 is used to analyze the speed sequence for non-motorized vehicles to determine whether the non-motorized vehicles have engaged in speeding behavior on curves.

[0088] It should be noted that the specific manner in which each module performs its operation in the apparatus described in the above embodiments has been described in detail in the embodiments of the method, and will not be elaborated here.

[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0090] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying non-motorized vehicle speeding based on image sequences, characterized in that, include: Acquire image sequence data of non-motorized vehicles in low-light curved passages; Based on the image sequence data, non-motorized vehicle boundary features and channel structure parameters are extracted, and based on the extracted features and channel structure parameters, mapping processing is performed to obtain the set of embedded trajectory points of non-motorized vehicles in the curved channel; The embedded trajectory point set is processed by constructing a direction-sensitive tensor graph and modeling a multi-scale steering perception graph to obtain a multi-dimensional graph structure containing information on time dimension, spatial direction and steering curvature. The multidimensional graph structure is processed by an angular graph attention mechanism and a convolutional trajectory transformation network to obtain a speed estimation vector sequence for non-motorized vehicles; The velocity estimation vector sequence is subjected to acceleration filtering and compensation processing to obtain a velocity sequence with the turning inertia error removed; The speed sequence is analyzed to determine whether the non-motorized vehicle is speeding on curves.

2. The non-motorized vehicle speeding identification method based on image sequences according to claim 1, characterized in that, Based on the image sequence data, non-motorized vehicle boundary features and channel structure parameters are extracted, and mapping processing is performed based on the extracted features and channel structure parameters, including: The image sequence data is processed by boundary tensor response extraction. By constructing a multi-frequency guided edge tensor response model of the non-motorized vehicle candidate region in each image frame of the image sequence data, a preliminary non-motorized vehicle boundary response map is obtained. The preliminary non-motorized vehicle boundary response map is subjected to local boundary curvature estimation. By performing multi-scale orientation fitting on the non-motorized vehicle boundary points and introducing a second-order guidance term in the strong turning region to suppress image noise interference, a set of boundary curvature points is obtained. The boundary curvature point set is subjected to channel structure-guided perspective geometry modeling. By utilizing the channel structure parameters of the image sequence data, the trajectory in the image frame is mapped from the pixel domain to the curvature space in a multi-stage manner to obtain the predicted motion path sequence of the non-motorized vehicle. The motion path sequence is processed by manifold embedding to obtain the set of embedded trajectory points of non-motorized vehicles in the curved passage.

3. The non-motorized vehicle speeding identification method based on image sequences according to claim 1, characterized in that, The embedded trajectory point set is subjected to orientation-sensitive tensor graph construction and multi-scale steering-aware graph modeling, including: The embedded trajectory point set is subjected to local direction encoding processing, wherein the motion direction between adjacent trajectory points is represented as a directed edge in a tensor to obtain an initial direction tensor graph. The initial orientation tensor is subjected to spatial orientation consistency optimization processing, wherein the spatial neighborhood average field is introduced into the initial orientation tensor and it is smoothed and irregularly turned to change the region to obtain the optimized tensor structure. The optimized tensor graph structure is subjected to multi-scale steering response modeling. By performing tensor domain hierarchical modeling on the steering gradient information at different scales, the feature response corresponding to the preset steering mode is extracted to obtain the steering perception graph structure. The steering perception graph structure is subjected to multidimensional graph structure fusion processing. By fusing the trajectory timestamp, position coordinates, direction vector and local curvature encoding into high-order graph node features, a multidimensional graph structure containing time dimension, spatial direction and steering curvature information is obtained.

4. The non-motorized vehicle speeding identification method based on image sequences according to claim 1, characterized in that, The multidimensional graph structure is processed through an angular graph attention mechanism and a convolutional trajectory transformation network, including: The multidimensional graph structure is subjected to attention weight initialization processing guided by angular partial derivative. The angular offset of each node in the multidimensional graph structure within its directional neighborhood is used as the kernel weight value for attention calculation to obtain the graph attention initial weight matrix. The initial weight matrix of the graph attention is subjected to node neighborhood feature aggregation processing. By fusing direction, time and curvature information in the variable neighborhood of the node, an aggregated feature tensor is formed, and a set of node state vectors is obtained. The set of node state vectors is subjected to trajectory convolution transformation. By performing sliding window convolution operation based on direction sequence on the sequential time axis of the trajectory, and introducing channel curvature as a convolution weight adjustment factor, a convolution feature map reflecting the dynamics of the trajectory time series is obtained. The convolutional feature map is subjected to velocity vector analysis. By performing first and second derivative analysis on the node sequence in the convolutional output domain, the velocity estimation vector sequence of non-motorized vehicles is obtained.

5. The non-motorized vehicle speeding identification method based on image sequences according to claim 1, characterized in that, The velocity estimation vector sequence is subjected to acceleration filtering and compensation processing, including: The velocity estimation vector sequence is subjected to local acceleration response extraction processing. By calculating the acceleration at each time node in the velocity estimation vector sequence and using the direction switching rate as a suppression factor, the original acceleration vector sequence of the non-motorized vehicle is obtained. The original acceleration vector sequence is subjected to inertial disturbance identification processing. By setting a steering inertial detection interval within the high curvature segment of the embedded trajectory, the non-real velocity fluctuation region dominated by the turning trajectory structure is identified, and a set of inertial disturbance time period markers is obtained. The set of inertial disturbance time period markers is reconstructed using a speed smoothing sequence. By weighting the speed values ​​within the marked time periods, a preliminary speed smoothing sequence for non-motorized vehicles is obtained. The initial velocity smoothing sequence is subjected to global trend calibration, wherein the overall velocity curve is refitted under the premise of ensuring energy conservation to obtain a velocity sequence with the turning inertia error removed.

6. A non-motorized vehicle speeding recognition device based on image sequences, characterized in that, include: The acquisition unit is used to acquire image sequence data of non-motorized vehicles in low-light curved passages; The mapping unit is used to extract non-motorized vehicle boundary features and channel structure parameters based on the image sequence data, and to perform mapping processing based on the extracted features and channel structure parameters to obtain the set of embedded trajectory points of non-motorized vehicles in the curved channel; The modeling unit is used to construct a direction-sensitive tensor graph and perform multi-scale steering perception graph modeling processing on the embedded trajectory point set to obtain a multi-dimensional graph structure containing time dimension, spatial direction and steering curvature information. The processing unit is used to process the multidimensional graph structure through an angular graph attention mechanism and a convolutional trajectory transformation network to obtain a speed estimation vector sequence for non-motorized vehicles; The removal unit is used to perform acceleration filtering and compensation processing on the velocity estimation vector sequence to obtain a velocity sequence with the turning inertia error removed. The analysis unit is used to analyze the speed sequence for non-motorized vehicles to determine whether they are engaging in speeding behavior on curves.

7. The non-motorized vehicle speeding recognition device based on image sequence according to claim 6, characterized in that, The mapping unit includes: The first mapping subunit is used to perform boundary tensor response extraction processing on the image sequence data. By constructing a multi-frequency guided edge tensor response model of the non-motor vehicle candidate region in each image frame of the image sequence data, a preliminary non-motor vehicle boundary response map is obtained. The second mapping subunit is used to perform local boundary curvature estimation processing on the preliminary non-motorized vehicle boundary response map. By performing multi-scale orientation fitting on the non-motorized vehicle boundary points and introducing a second-order guidance term in the strong turning region to suppress image noise interference, a set of boundary curvature points is obtained. The third mapping subunit is used to perform channel structure-guided perspective geometry modeling on the boundary curvature point set. By using the channel structure parameters of the image sequence data to perform multi-stage mapping from the pixel domain to the curvature space on the trajectory in the image frame, the estimated motion path sequence of the non-motorized vehicle is obtained. The fourth mapping subunit is used to perform manifold embedding processing on the motion path sequence to obtain the set of embedded trajectory points of non-motorized vehicles in the curve channel.

8. The non-motorized vehicle speeding recognition device based on image sequence according to claim 6, characterized in that, The modeling unit includes: The first modeling subunit is used to perform local direction encoding processing on the embedded trajectory point set, wherein the motion direction between adjacent trajectory points is represented as a directed edge in a tensor to obtain an initial direction tensor graph. The second modeling subunit is used to perform spatial orientation consistency optimization processing on the initial orientation tensor graph. Specifically, the orientation average field in the spatial neighborhood is introduced into the initial orientation tensor graph and it is smoothly and irregularly turned to change the region to obtain the optimized tensor graph structure. The third modeling subunit is used to perform multi-scale steering response modeling on the optimized tensor graph structure. By performing tensor domain hierarchical modeling on the steering gradient information at different scales, the feature response corresponding to the preset steering mode is extracted to obtain the steering perception graph structure. The fourth modeling subunit is used to perform multi-dimensional graph structure fusion processing on the steering perception graph structure. By fusing the trajectory timestamp, position coordinates, direction vector and local curvature encoding into high-order graph node features, a multi-dimensional graph structure containing time dimension, spatial direction and steering curvature information is obtained.

9. The non-motorized vehicle speeding recognition device based on image sequence according to claim 6, characterized in that, The processing unit includes: The first processing subunit is used to perform attention weight initialization processing guided by angular partial derivative on the multidimensional graph structure. By using the angular offset of each node in the multidimensional graph structure in its directional neighborhood as the kernel weight value for attention calculation, the graph attention initial weight matrix is ​​obtained. The second processing subunit is used to perform node neighborhood feature aggregation processing on the graph attention initial weight matrix. By fusing direction, time and curvature information in the variable neighborhood of the node, an aggregated feature tensor is formed to obtain a set of node state vectors. The third processing subunit is used to perform trajectory convolution transformation on the set of node state vectors. By performing sliding window convolution operation based on direction sequence on the sequential time axis of the trajectory, and introducing channel curvature as a convolution weight adjustment factor, a convolution feature map reflecting the dynamics of the trajectory time series is obtained. The fourth processing subunit is used to perform velocity vector analysis on the convolutional feature map. By performing first-order and second-order derivative analysis on the node sequence in the convolutional output domain, the velocity estimation vector sequence of the non-motorized vehicle is obtained.

10. The non-motorized vehicle speeding recognition device based on image sequence according to claim 6, characterized in that, The removal unit includes: The first removal subunit is used to perform local acceleration response extraction processing on the velocity estimation vector sequence. By calculating the acceleration at each time node in the velocity estimation vector sequence and using the direction switching rate as a suppression factor, the original acceleration vector sequence of the non-motorized vehicle is obtained. The second removal subunit is used to perform inertial disturbance identification processing on the original acceleration vector sequence. By setting a steering inertial detection interval in the high curvature segment of the embedded trajectory, the non-real speed fluctuation area dominated by the turning trajectory structure is identified, and a set of inertial disturbance time period markers is obtained. The third removal subunit is used to perform speed smoothing sequence reconstruction processing on the inertial disturbance time period mark set. By weighting the speed values ​​within the mark period, a preliminary speed smoothing sequence of non-motorized vehicles is obtained. The fourth removal subunit is used to perform global trend calibration processing on the preliminary velocity smoothing sequence, wherein the overall velocity curve is refitted under the premise of ensuring energy conservation to obtain a velocity sequence with the turning inertia error removed.

Images