A method and system for intelligent driving warning

Abstract

The application belongs to the technical field of intelligent driving, and discloses a kind of for intelligent driving early warning method and system;Method includes: real-time detection road both sides stationary vehicle, and through the space arrangement relationship between adjacent vehicle edges and pixel spacing, extract N crossable gap region;Based on the image brightness variation features extracted in N crossable gap region, identify M crossable gap region caused by light and shadow disturbance of target pedestrian activity;Based on the corresponding light and shadow disturbance feature set extracted in M crossable gap region, identify Q crossable gap region corresponding to the trend of pedestrian crossing forward, and evaluate the corresponding danger state level in combination with the current position and vehicle current speed of the vehicle driving;Based on the danger state level, execute danger state classification adaptive warning strategy;The application realizes the risk prediction and classification adaptive warning of pedestrian sudden crossing in the non-line-of-sight complex occlusion traffic scene.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving technology, and more specifically, to a method and system for early warning in intelligent driving. Background Technology

[0002] With the rapid development of intelligent driving technology, pedestrian detection and collision warning systems based on multi-sensor perception have become an important component of intelligent driving assistance. Existing intelligent driving warning methods typically rely on onboard cameras, millimeter-wave radar, or lidar to directly identify pedestrian targets and determine whether a pedestrian has entered the vehicle's path using target detection boxes or contour trajectories. In open roads or unobstructed environments, such methods can effectively identify pedestrians and issue early warnings. However, in typical non-line-of-sight occlusion traffic scenarios such as urban roads, parking lots, and narrow streets where there are many stationary vehicles obstructing the view, pedestrians often suddenly appear within a passable gap in a short period of time. This results in traditional warning methods based on explicit target detection lacking timeliness and accuracy in identifying sudden pedestrian crossings, posing potential safety hazards.

[0003] However, in practical applications, existing technologies lack systematic modeling of the dynamic changing trends of disturbance areas and the spatial relationship of gap areas, making it difficult to accurately determine whether disturbances have a tendency to cross the lane. Furthermore, existing warning strategies often employ fixed thresholds or single braking trigger logic, failing to combine disturbance prediction trajectories, vehicle braking capabilities, and driving status for multi-level hazard classification responses. This results in high false alarm rates or untimely interventions, making it impossible to achieve high-confidence identification and adaptive warnings in complex, obscured environments.

[0004] In view of this, the present invention proposes a method and system for early warning in intelligent driving to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution: a method for early warning in intelligent driving, comprising:

[0006] Real-time detection of stationary vehicles on both sides of the road, and extraction of N traversable gap regions based on the spatial arrangement relationship and pixel spacing between the edges of adjacent vehicles;

[0007] Based on the image brightness change features extracted from N traversable gap regions, M traversable gap regions caused by the light and shadow disturbances caused by the activity of the target pedestrian are identified.

[0008] Based on the extraction of corresponding light and shadow disturbance feature sets within M traversable gap regions, Q traversable gap regions corresponding to pedestrians with a forward crossing trend are identified. Combined with the current position and speed of the vehicle, the corresponding danger level is assessed for the Q traversable gap regions.

[0009] Based on the hazard status levels of Q traversable gap regions, an adaptive early warning strategy for hazard status classification is implemented.

[0010] Furthermore, based on the hazard level of Q traversable gap regions, the method for implementing an adaptive early warning strategy for hazard level classification includes:

[0011] Obtain the corresponding danger level for each of the Q traversable gap regions;

[0012] When the danger level is high, the system generates an emergency audible and visual alarm signal, which warns the driver through the dashboard warning lights, buzzer, or vehicle head-up display. Based on the safe distance threshold corresponding to the high danger level, the system calculates the vehicle's target deceleration speed and automatically triggers deceleration control to adjust the vehicle's current speed to the target deceleration speed.

[0013] When the danger level is moderate, the system generates a warning signal to prompt the driver to increase attention and prepare to slow down through visual or auditory means.

[0014] When the hazard level is low, the system continues to monitor the traversable gap area.

[0015] Furthermore, the method for obtaining the light and shadow perturbation feature set includes:

[0016] The continuous image corresponding to each of the M traversable gap regions is divided into multiple image frames. The active perturbation regions corresponding to adjacent image frames are extracted. The edge contour of each active perturbation region is extracted to generate the minimum bounding rectangle and obtain the corresponding centroid coordinates. Based on the centroid coordinates of adjacent image frames, the light and shadow expansion direction vector and the perturbation centroid movement velocity are constructed. The light and shadow expansion direction vector is added to the light and shadow expansion direction vector set, and the perturbation centroid movement velocity is added to the perturbation centroid movement velocity set until all image frames are processed. The light and shadow expansion direction vector set and the perturbation centroid movement velocity set are used to construct the corresponding light and shadow perturbation feature set.

[0017] Furthermore, the method for obtaining the active disturbance region includes:

[0018] Obtain the average brightness change amplitude and brightness variance increment corresponding to the image frames in the traversable gap region;

[0019] Pixels in the image frame of the traversable gap region whose average brightness change is greater than a preset average brightness change threshold and whose brightness variance increment is greater than a preset brightness variance increment threshold are marked as active perturbation pixels.

[0020] All active perturbation pixels are used to construct an active perturbation region.

[0021] Furthermore, the method for identifying the Q traversable gap regions corresponding to the pedestrian's forward traversing trend includes:

[0022] Based on the light and shadow perturbation feature set of each of the M traversable gap regions, the set of light and shadow expansion direction vectors and the set of perturbation centroid movement velocity are extracted respectively;

[0023] The average light and shadow expansion direction vector is calculated based on the set of light and shadow expansion direction vectors, the average perturbation centroid movement velocity is calculated based on the set of perturbation centroid movement velocities, and the average light and shadow expansion direction vector is converted into light and shadow expansion direction angle.

[0024] The center coordinates are extracted based on the active disturbance region of the current image frame corresponding to each traversable gap region. The target pedestrian position in the future unit time is predicted by combining the light and shadow expansion direction angle with the center coordinates.

[0025] When the predicted position of the target pedestrian exceeds the boundary of the corresponding traversable gap area, the traversable gap area is determined to be a traversable gap area with a tendency for the pedestrian to cross forward; finally, Q traversable gap areas caused by the light and shadow disturbance of the target pedestrian activity are obtained.

[0026] Furthermore, the assessment method for the hazard level corresponding to the Q traversable gap regions includes:

[0027] For each traversable gap area with a pedestrian tendency to cross forward, obtain the corresponding center coordinates of the active disturbance area, calculate the intersection prediction distance based on the center coordinates of the active disturbance area and the current position of the vehicle, calculate the emergency braking distance based on the current speed of the vehicle, add the emergency braking distance to the preset safety redundancy distance threshold one to obtain safety distance threshold one, and add the emergency braking distance to the preset safety redundancy distance threshold two to obtain safety distance threshold two.

[0028] When the predicted rendezvous distance is greater than or equal to the second safe distance threshold, the corresponding hazard level is set to low hazard level; when the predicted rendezvous distance is greater than or equal to the first safe distance threshold and less than the second safe distance threshold, the corresponding hazard level is set to medium hazard level; when the predicted rendezvous distance is less than the first safe distance threshold, the corresponding hazard level is set to high hazard level.

[0029] Furthermore, the method for extracting the image brightness variation features of N traversable gap regions includes:

[0030] The continuously acquired images of the traversable gap region are divided into T image frames. For each traversable gap region, the average brightness change amplitude and brightness variance increment are calculated sequentially for the adjacent t-th and t-1-th frames. The image pixel grayscale is divided into B equally spaced amplitude intervals, the number of pixels in each amplitude interval is counted, and the corresponding brightness perturbation entropy is calculated. The average brightness change amplitude, brightness variance increment, and brightness perturbation entropy are added to the corresponding sets. After all image frames are processed, each set is used to construct the image brightness change feature corresponding to the traversable gap region. This process is repeated until all N traversable gap regions are processed, resulting in the image brightness change features of the N traversable gap regions.

[0031] Furthermore, the method for identifying the M traversable gap regions caused by the light and shadow disturbances resulting from the movement of the target pedestrian includes:

[0032] The image brightness change features corresponding to the N traversable gap regions are input into the pedestrian activity confidence evaluation model to obtain the corresponding pedestrian activity confidence. The traversable gap regions with pedestrian activity confidence greater than the preset pedestrian activity confidence threshold are determined as traversable gap regions of light and shadow disturbance caused by the target pedestrian activity, thus obtaining the M traversable gap regions of light and shadow disturbance caused by the target pedestrian activity.

[0033] Furthermore, the methods for obtaining the N traversable gap regions include:

[0034] The input image is used to extract stationary vehicle targets using an object detection algorithm, and the stationary vehicle region is constructed.

[0035] Sort the edge positions of adjacent vehicles in the stationary vehicle area and construct an edge arrangement sequence;

[0036] For each pair of adjacent vehicles in the edge arrangement sequence, the gap between adjacent vehicles is calculated as the pixel spacing of the gap in the input image. Then, combined with the camera intrinsic matrix, shooting height and vehicle forward IMU attitude parameters of the current input image, the pixel spacing is converted into the gap width value in the actual space using a monocular projection geometry model.

[0037] Compare all calculated gap width values ​​with a preset passable width threshold, and mark the gaps between adjacent vehicles with gap width values ​​greater than the passable width threshold as passable gap areas, thus obtaining N passable gap areas.

[0038] A warning system for intelligent driving, and a method for warning of intelligent driving, comprising:

[0039] The gap region recognition module is used to detect stationary vehicles on both sides of the road in real time, and extract N passable gap regions by the spatial arrangement relationship and pixel spacing between the edges of adjacent vehicles.

[0040] The activity feature diagnosis module identifies M traversable gap regions caused by the light and shadow disturbances resulting from the activity of a target pedestrian based on the image brightness change features extracted from N traversable gap regions.

[0041] The traversal risk assessment module extracts the corresponding light and shadow disturbance feature sets based on M traversable gap regions, identifies Q traversable gap regions corresponding to pedestrians with a forward traversal trend, and assesses the corresponding danger level for the Q traversable gap regions by combining the current position and speed of the vehicle.

[0042] The hazard warning module executes a hazard classification adaptive warning strategy based on the hazard levels of Q traversable gap areas.

[0043] Compared with existing technologies, the technical effects and advantages of the present invention for a method and system for early warning in intelligent driving are as follows:

[0044] This application constructs a hierarchical adaptive early warning method for predicting the risk of sudden pedestrian crossings in non-line-of-sight occluded traffic scenarios by deeply integrating the traversable gap regions formed between stationary vehicles on both sides of the road with light and shadow disturbance features. Specifically, the system first uses a gap region identification module to accurately acquire N traversable gap regions with pedestrian crossing potential based on the geometric relationship of adjacent vehicle edges, pixel spacing, and camera pose parameters, thus achieving explicit modeling of potential passage areas. Subsequently, by jointly extracting the brightness change amplitude, brightness variance increment, and brightness disturbance entropy of consecutive frames, the system accurately characterizes the locality, intensity, and spatial complexity of the disturbance region, effectively distinguishing between local pedestrian activity and global background fluctuations, and significantly reducing the false detection rate under complex lighting conditions.

[0045] This application further employs future position prediction based on the average light and shadow expansion direction and the movement speed of the disturbance centroid to determine whether the disturbance target has a tendency to cross the gap boundary. Compared with traditional recognition methods that rely on the exposure of the target outline, this method has stronger early perception capabilities and the ability to discriminate the dynamic behavior of occluded pedestrians. Multi-level safety thresholds are constructed based on the intersection prediction distance and emergency braking distance to achieve quantitative classification of different danger levels. Combined with the vehicle's current speed, the target deceleration speed is automatically calculated, and multi-level response strategies such as emergency braking, visual and auditory warnings, or continuous monitoring are triggered according to the risk level.

[0046] In summary, this application not only enables high-confidence identification of pedestrians based on weak light and shadow disturbances and regional expansion trends before they have fully entered the main lane, but also achieves autonomous vehicle deceleration and dynamic graded handling when the risk level increases. This improves the system's response timeliness and intervention accuracy in non-line-of-sight occluded "ghost pedestrian" situations. Compared with existing technical solutions that rely solely on visual target detection or fixed threshold alarms, this invention significantly improves the accuracy of pedestrian crossing risk identification, the adaptability of warning response, and the rationality of vehicle braking control in complex occlusion environments, demonstrating stronger practical value and engineering application prospects. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of an intelligent driving early warning system according to Embodiment 1 of the present invention;

[0048] Figure 2 This is a flowchart of a method for early warning in intelligent driving according to Embodiment 2 of the present invention;

[0049] Figure 3 Flowchart of the method for implementing the adaptive early warning strategy for hazardous state classification;

[0050] Figure 4 Flowchart of a method for identifying Q traversable gap regions with a pedestrian tendency to cross forward;

[0051] Figure 5 The flowchart shows the assessment method for the hazard level corresponding to Q traversable gap regions. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be described in detail, clearly, and completely below with reference to the accompanying drawings. It should be particularly noted that the specific embodiments described below are only for better illustrating and explaining the technical solutions of the present invention, and are intended to enable those skilled in the art to better understand and implement the present invention, and should not be construed as limiting the scope of protection of the present invention. Without departing from the spirit and substance of the present invention, those skilled in the art can modify, adjust, or make equivalent substitutions based on the content disclosed in the present invention, and these should all be considered within the scope of protection of the present invention.

[0053] Example 1:

[0054] Please see Figure 1 As shown, this embodiment discloses a warning system for intelligent driving, including a gap area recognition module, an activity feature diagnosis module, a crossing risk assessment module, and a dangerous state warning module. Each module is connected by wires and / or wirelessly to realize data transmission.

[0055] To further clarify the technical problem to be solved by this application and its background, before proceeding with specific implementation methods, the relevant reaction mechanism, the limitations of the prior art, and the practical difficulties faced by those skilled in the art in solving the problem will be explained in detail.

[0056] Specifically, in the context of intelligent driving early warning applications, pedestrians suddenly rushing out between vehicles constitute a typical complex traffic scenario involving non-line-of-sight occlusion, a technical challenge and safety pain point commonly faced by existing intelligent perception systems. Specifically, in typical traffic environments such as urban roads, school perimeters, and residential entrances, numerous temporarily or permanently parked stationary vehicles often exist on both sides of the road. These vehicles form a continuous occlusion strip in space, obscuring the direct line of sight between the driver's sensors (including forward-facing cameras, millimeter-wave radar, and lidar) and potential pedestrians crossing the road. Before a potential pedestrian enters the main lane, their position is entirely in a "non-line-of-sight" state, making them impossible to directly observe or detect.

[0057] However, due to the gaps between vehicles, obscured pedestrians have a high probability of suddenly crossing the lane at uncertain speeds, easily leading to sudden collisions caused by unexpected pedestrian movements. Traditional methods based on target bounding box detection or forward-view image semantic recognition typically only produce an effective response after part or all of the target is exposed to the field of view, often resulting in delayed hazard identification and failing to meet the actual needs of advance warning and pedestrian trend perception.

[0058] Therefore, determining the potential for pedestrian crossing behavior based on indirect clues such as environmental structure, light and shadow disturbances, or reflection changes when the target pedestrian has not yet fully revealed their form and is still within the vehicle's obstruction area has become one of the key technical problems that current intelligent driving warning systems urgently need to solve. This application addresses this type of high-risk scenario with non-line-of-sight obstruction by identifying gap areas with potential for passage and further establishing a joint analysis model of dynamic disturbances and target trends. This enables early prediction and graded response to the risk of pedestrians rushing out from obstructed areas, significantly improving the system's warning timeliness and proactive decision-making in high-risk non-line-of-sight obstruction scenarios.

[0059] The gap region recognition module is used to detect stationary vehicles on both sides of the road in real time, and extract N passable gap regions by the spatial arrangement relationship and pixel spacing between the edges of adjacent vehicles.

[0060] Methods for obtaining N traversable gap regions include:

[0061] The input image is used to extract stationary vehicle targets using an object detection algorithm, and the stationary vehicle region is constructed.

[0062] Sort the edge positions of adjacent vehicles in the stationary vehicle area and construct an edge arrangement sequence;

[0063] For each pair of adjacent vehicles in the edge arrangement sequence, the gap between adjacent vehicles is calculated as the pixel spacing of the gap in the input image. Then, combined with the camera intrinsic matrix, shooting height and vehicle forward IMU attitude parameters of the current input image, the pixel spacing is converted into the gap width value in the actual space using a monocular projection geometry model.

[0064] Compare all calculated gap width values ​​with a preset passable width threshold, and mark the gaps between adjacent vehicles with gap width values ​​greater than the passable width threshold as passable gap areas, thus obtaining N passable gap areas.

[0065] It should be noted that, in a preferred embodiment of this application, a convolutional neural network with multi-scale perception capability, such as YOLOv5 or SSD, is preferably used to identify candidate regions in the input image that belong to the category of stationary vehicles. To ensure the accuracy of the recognition results, the system employs a temporal stability determination mechanism to check the positional consistency of vehicle detection boxes in multiple consecutive frames of input images. If the positional information of the same target does not drift within a preset time window and the velocity estimation is close to zero, it is marked as a stationary vehicle.

[0066] The passable width threshold can be set between 0.5 meters and 0.8 meters to ensure that the lateral space required for normal passage of adult pedestrians is covered.

[0067] The camera intrinsic parameter matrix is:

[0068] Where NC is the camera intrinsic parameter matrix, f x f is the equivalent focal length in the horizontal direction. y The equivalent focal length in the vertical direction is (c x c y The principal point (c) is the pixel coordinate of the principal point in the image coordinate system. x c y The term "pinhole projection" refers to the position of the actual intersection point of the camera's optical axis and the image sensor's imaging plane in the pixel coordinate system. The camera's optical axis is the optical center ray of the lens. The camera's intrinsic parameter matrix is ​​used to describe the pinhole projection relationship between pixel coordinates and camera coordinates.

[0069] The vehicle's forward IMU attitude parameters refer to the pitch angle, roll angle, and yaw angle, which are output in real time by the inertial measurement unit (IMU) installed on the vehicle. Among them, the pitch angle and roll angle determine the degree of tilt of the camera's optical axis relative to the ground, while the yaw angle describes the horizontal angle between the vehicle's head direction and the X-axis of the world coordinate system. In this application specification, the X-axis of the world coordinate system refers to the lateral reference direction in a three-dimensional right-handed reference coordinate system that is fixed relative to the ground.

[0070] The monocular projection geometry model is based on a pinhole imaging model and external camera mounting parameters. It calculates the projected position of pixels in the input image in actual space using the camera's intrinsic parameter matrix, shooting height, and vehicle forward IMU attitude parameters. For pixel pairs located at the edge of gaps in the input image, those skilled in the art can solve for the lateral coordinate difference of the pixel pair on the ground plane using a ray-plane intersection method, thereby obtaining the gap width between adjacent vehicles. This conversion process belongs to the image-physical scale mapping method already mastered by those skilled in the art, with clear mathematical principles and engineering implementation paths. Various parameters can also be configured using visual calibration tools, such as OpenCV and the MATLAB vision toolbox.

[0071] Furthermore, in a preferred embodiment of this application, the purpose of the N traversable gap regions obtained by the gap region identification module is to identify gap regions with potential "pedestrian crossing probability" from stationary vehicles parked on both sides of the road during vehicle travel, and to provide clear spatial positioning basis for subsequent human target activity identification and traversal risk trend judgment based on image brightness perturbation features, thus ensuring at the geometric level that the image region on which subsequent detection depends has clear physical scale semantics.

[0072] Specifically, the gaps formed between vehicles at the road edge are often blind spots, especially at intersections of sidewalks and roadways where pedestrians can easily dart across. The system detects stationary vehicle edges and calculates the pixel spacing of the gaps, further combining camera imaging parameters and vehicle posture information to convert the pixel width of the gaps on the image plane into the physical width in real space. This allows the system to filter out areas that meet the conditions for human passage, avoiding misclassification of closed structures or gaps caused solely by visual overlap as traversable areas.

[0073] In summary, this embodiment, by acquiring N traversable gap regions, essentially achieves explicit modeling and screening of the spatial preconditions for possible pedestrian crossing behavior in static occlusion scenarios. This provides a precise, constrained, and physically measurable foundation for subsequent dynamic target recognition and crossing risk prediction, significantly enhancing the adaptability and recognition accuracy of the intelligent driving early warning system in typical complex traffic scenarios with non-line-of-sight occlusion. Non-line-of-sight occlusion refers to situations where a target pedestrian is obscured by an intermediate obstacle at the current time, not appearing in the directly visible image or effective radar reflection range, but indirect features such as the pedestrian's image, reflection, shadow, and light and shadow disturbances can be observed, or the pedestrian's dynamic movement trend can be inferred from the occlusion edge.

[0074] The activity feature diagnosis module identifies M traversable gap regions caused by the light and shadow disturbances resulting from the activity of a target pedestrian, based on the image brightness change features extracted from N traversable gap regions.

[0075] Methods for extracting image brightness variation features from N traversable gap regions include:

[0076] S100: Divide the continuously acquired images corresponding to the traversable gap region into T traversable gap region image frames; T is a positive integer; let the initial value of t be 2, and the value of t ranges from 2 to T; let the initial value of n be 1, and the initial value of n ranges from 1 to N; let the initial value of the counter variable JS be 1.

[0077] S101: Based on the image frames of the tth and t-1th traversable gap regions corresponding to the nth traversable gap region, calculate the average brightness change amplitude between the image frames of the tth and t-1th traversable gap regions corresponding to the nth traversable gap region, and denoted as the JSth average brightness change amplitude.

[0078] Based on the JSth average brightness change amplitude and the tth traversable gap image frame of the nth traversable gap region, the brightness variance increment between the tth and t-1th traversable gap image frames corresponding to the nth traversable gap region is calculated and denoted as the JSth brightness variance increment.

[0079] The pixel grayscale depth of the image acquisition device is divided into B equally spaced amplitude intervals; the number of pixels in each equally spaced amplitude interval of the nth traversable gap region is counted and denoted as the number of pixels in the amplitude interval; based on the number of pixels in the B amplitude intervals, the luminance perturbation entropy corresponding to the tth traversable gap region image frame of the nth traversable gap region is calculated and denoted as the JSth luminance perturbation entropy.

[0080] It should be noted that the pixel grayscale depth of the image acquisition device ranges from [L] to [L]. min L max ], L min L represents the minimum pixel grayscale depth of an image acquisition device. max This represents the maximum value of the pixel grayscale depth of the image acquisition device. For example, the pixel grayscale depth of the image acquisition device is usually [-255, 255].

[0081] S102: Add the JSth average brightness change amplitude to the set of average brightness change amplitudes corresponding to the nth passable gap region, add the JSth brightness variance increment to the set of brightness variance increments corresponding to the nth passable gap region, and add the JSth brightness disturbance entropy to the set of brightness disturbance entropy corresponding to the nth passable gap region.

[0082] S103: Let t = t + 1. If t is less than or equal to T, then let JS = JS + 1 and return to S101 to continue execution. If t is greater than T, then construct the set of average brightness change amplitude, the set of brightness variance increment, and the set of brightness perturbation entropy into the image brightness change feature corresponding to the nth traversable gap region, and execute S104.

[0083] S104: Let n = n + 1. If n is less than or equal to N, then let t = 2, let JS = 1, and return to S101 to continue execution; if n is greater than N, then obtain the image brightness change features corresponding to N traversable gap regions, and end the current process.

[0084] The method for calculating the average brightness variation includes:

[0085]

[0086] in, NUM represents the average brightness change between the t-th and (t-1)-th image frames corresponding to the n-th traversable gap region. n This represents the total number of pixels contained in the nth traversable gap region. This represents the pixel brightness corresponding to the i-th pixel in the image frame of the t-th traversable gap region of the n-th traversable gap region. This represents the brightness of the i-th pixel in the image frame of the (t-1)-th traversable gap region of the n-th traversable gap region, where i is the index variable of the summation formula.

[0087] The average brightness change amplitude is a core quantitative indicator used to measure the overall light and shadow disturbance intensity of the nth traversable gap region between two consecutive image frames. Specifically, by averaging the brightness difference of each pixel within the traversable gap region, the overall brightness jump of the traversable gap region within the current adjacent traversable gap region image frame can be intuitively reflected. When the average brightness change amplitude is higher than the brightness change amplitude threshold, it is determined that there is significant light and shadow disturbance within the traversable gap region image frame, which is a priori signal that a potential pedestrian is active while crossing the target.

[0088] The method for calculating the luminance variance increment includes:

[0089]

[0090] in, This represents the luminance variance increment between the t-th and (t-1)-th image frames corresponding to the n-th traversable gap region. The luminance variance increment reflects the spatial dispersion of luminance changes within the traversable gap region; a larger luminance variance increment indicates higher spatial non-uniformity of luminance disturbances within the traversable gap region.

[0091] Specifically, when the increase in brightness variance increases, it indicates that the brightness change is concentrated in a local area, which can be determined as light and shadow disturbance caused by activities such as pedestrians blocking the light source or moving projections; conversely, if the brightness change is relatively uniform, that is, the increase in brightness variance does not change significantly, it is determined to be a change in the overall ambient lighting or automatic exposure adjustment of the camera, and there is no potential pedestrian crossing the target.

[0092] Furthermore, when background interference factors such as swaying tree shadows, flickering large-area advertising screens, or reflective fluctuations on vehicle surfaces are present, these interference factors often cause brightness changes in a relatively uniform manner throughout the entire passable gap area, meaning that the brightness of each pixel experiences small, synchronous fluctuations of approximately the same amplitude. By using the average brightness change amplitude and the brightness variance increment as joint features, this application can effectively distinguish between local spatial disturbances caused by target pedestrians and overall light and shadow fluctuations caused by environmental factors. This joint feature discrimination mechanism effectively suppresses false triggering in complex backgrounds and improves the stability and practicality of the warning system in non-line-of-sight occlusion environments.

[0093] The method for calculating the brightness perturbation entropy includes:

[0094]

[0095] in, Let SL be the luminance perturbation entropy corresponding to the t-th traversable gap region image frame of the n-th traversable gap region. b Let be the number of pixels in the amplitude interval corresponding to the b-th equally spaced amplitude interval, and b be the index variable of the summation formula.

[0096] It should be noted that, in a preferred embodiment of this application, the luminance perturbation entropy is used as an important discriminant index to reflect the distribution characteristics of luminance changes within the traversable gap area. Its core function is to measure the spatial complexity and amplitude dispersion of light and shadow perturbations within the traversable gap area, thereby distinguishing between changes caused by local occlusion of the target pedestrian and changes caused by overall ambient illumination fluctuations or random noise.

[0097] Specifically, when the brightness abruptly changes only in a localized area of ​​the gap, such as when a pedestrian partially peeks out from the gap, the pixel brightness difference is mainly concentrated in a few high-amplitude ranges, resulting in a lower brightness perturbation entropy value. Conversely, when the perturbation originates from a global or uniform lighting change, such as a large-scale shadow movement or exposure drift, the pixel brightness difference is distributed across multiple amplitude ranges, leading to a higher brightness perturbation entropy value. By combining this with the average brightness change amplitude and the brightness variance increment, the brightness perturbation entropy can be used to distinguish between localized, high-contrast perturbations (i.e., high-confidence pedestrian activity) and global, low-contrast perturbations (i.e., low-confidence environmental noise), significantly improving the system's accuracy in recognizing occluded pedestrian dynamics.

[0098] The methods for identifying M traversable gap regions caused by light and shadow disturbances due to the movement of a target pedestrian include:

[0099] S200: Let the initial value of n be 1, and the range of n is from 1 to N;

[0100] S201: Input the image brightness change features corresponding to the nth passable gap region into the pedestrian activity confidence evaluation model to obtain the corresponding pedestrian activity confidence.

[0101] If the confidence level of pedestrian activity is greater than the preset confidence level threshold of pedestrian activity, then the nth traversable gap region is marked as the traversable gap region of light and shadow disturbance caused by the target pedestrian activity.

[0102] S202: Let n = n + 1. If n is less than or equal to N, then return to S201 to execute. If n is greater than N, then the number of traversable gap regions marked as light and shadow disturbances caused by the target pedestrian activity is recorded as M. That is, the M traversable gap regions of light and shadow disturbances caused by the target pedestrian activity are obtained, and the current process ends.

[0103] The training method for the activity confidence assessment model includes:

[0104] An activity confidence assessment dataset is pre-constructed, comprising ZXD group activity confidence assessment data and corresponding pedestrian activity confidence scores, where ZXD is a positive integer. The activity confidence assessment data includes image brightness variation features, which include a set of average brightness variation amplitudes, a set of brightness variance increments, and a set of brightness perturbation entropies. The activity confidence assessment dataset is divided into a training set and a validation set. The training set is used for learning the parameters of the activity confidence assessment model, while the validation set is used to monitor the generalization performance and overfitting degree of the activity confidence assessment model in real time.

[0105] A deep neural network with a multilayer perceptron as its core is used as the activity confidence assessment model. The activity confidence assessment data is standardized and vectorized before being input into the deep neural network, which consists of an input layer, hidden layers, and an output layer. Each hidden layer uses a nonlinear activation function to extract features, and the output layer uses a softmax activation function to obtain the probability distribution corresponding to each pedestrian activity confidence. Finally, the pedestrian activity confidence corresponding to the highest probability is taken as the prediction result of the activity confidence assessment model. During training, the cross-entropy loss function is used as the optimization objective, and a gradient descent-type optimization algorithm is used to update the network weights. An early stopping strategy is set: when the prediction accuracy on the validation set reaches or exceeds a preset threshold, the activity confidence assessment model is determined to have converged and training is terminated.

[0106] It should be noted that the pedestrian activity confidence score represents the probability that there is an obstructed pedestrian activity within the currently traversable gap area, and its value ranges from [0, 1]. The closer the pedestrian activity confidence score is to 1, the higher the confidence that pedestrian activity disturbance has occurred within the corresponding traversable gap area; the closer the value is to 0, the more likely no pedestrian activity or disturbance has been detected in the corresponding area. In this embodiment, the pedestrian activity confidence score threshold can be set to [0.7, 0.85]. For example, in this application, the pedestrian activity confidence score threshold can be set to 0.8. Furthermore, the system can also be dynamically adjusted according to road traffic complexity, lighting conditions, or historical recognition performance to adapt to risk identification needs under different operating conditions.

[0107] The traversal risk assessment module extracts corresponding light and shadow disturbance feature sets from M traversable gap regions, identifies Q traversable gap regions corresponding to pedestrians' forward traversal tendency, and assesses the corresponding hazard level for the Q traversable gap regions based on the current position and speed of the vehicle. The hazard level includes low hazard, medium hazard, and high hazard.

[0108] The method for obtaining the light and shadow perturbation feature set includes:

[0109] S300: Let m be initially set to 1, and the value of m ranges from 1 to M; divide the continuously acquired images corresponding to the traversable gap region into T traversable gap region image frames; T is a positive integer; let t be initially set to 2, and the value of t ranges from 2 to T; let the initial value of the counting variable JM be 1;

[0110] S301: Extract the active disturbance region corresponding to the t-th traversable gap region image frame of the m-th traversable gap region, denoted as the first active disturbance region. Extract the edge contour of the first active disturbance region to obtain the edge contour of the first active disturbance region, and construct the minimum bounding rectangle of the edge contour of the first active disturbance region, denoted as the first bounding rectangle. Mark the center coordinates of the first bounding rectangle as the first centroid.

[0111] Extract the active perturbation region corresponding to the (t-1)th traversable gap region image frame of the mth traversable gap region, and denote it as the second active perturbation region. Extract the edge contour of the second active perturbation region to obtain the edge contour of the second active perturbation region, and construct the minimum bounding rectangle of the edge contour of the second active perturbation region, and denote it as the second bounding rectangle. Mark the center coordinates of the second bounding rectangle as the second centroid.

[0112] S302: Based on the first centroid and the second centroid, construct the JMth light and shadow expansion direction vector of the mth traversable gap region; based on the time interval between the first centroid, the second centroid and the image frames of the tth and t-1th traversable gap regions, construct the JMth perturbation centroid movement velocity of the mth traversable gap region.

[0113] S303: Add the JMth light and shadow expansion direction vector to the set of light and shadow expansion direction vectors corresponding to the mth traversable gap region, and add the JMth disturbance centroid movement speed to the set of disturbance centroid movement speeds corresponding to the mth traversable gap region.

[0114] S304: Let t = t + 1. If t is less than or equal to T, let JM = JM + 1 and return to S301 to continue execution. If t is greater than T, construct the light and shadow expansion direction vector set and the perturbation centroid movement velocity set into the light and shadow perturbation feature set corresponding to the mth traversable gap region and execute S305.

[0115] S305: Let m = m + 1. If m is less than or equal to M, then let t = 2, let JM = 1, and return to S301 to continue execution; if m is greater than M, then obtain the set of light and shadow disturbance features corresponding to the M traversable gap regions, and end the current process.

[0116] The method for constructing the light and shadow expansion direction vector includes:

[0117]

[0118] in, Let JM be the light and shadow expansion direction vector of the m-th traversable gap region. Let x be the x-coordinate of the first centroid of the m-th traversable gap region. Let x be the x-coordinate of the second centroid of the m-th traversable gap region. Let be the ordinate of the first centroid of the m-th traversable gap region. The ordinate corresponding to the second centroid of the m-th traversable gap region.

[0119] The method for constructing the velocity of the disturbed centroid includes:

[0120]

[0121] in, Let JM be the moving velocity of the JM-th perturbation centroid in the m-th traversable gap region, and ΔSJJG be the time interval between the t-th and t-1-th traversable gap region image frames.

[0122] The method for obtaining the active disturbance region includes:

[0123] Obtain the average brightness change amplitude and brightness variance increment corresponding to the image frames in the traversable gap region;

[0124] Pixels in the image frame of the traversable gap region whose average brightness change is greater than a preset average brightness change threshold and whose brightness variance increment is greater than a preset brightness variance increment threshold are marked as active perturbation pixels.

[0125] All active perturbation pixels are used to construct an active perturbation region.

[0126] It should be noted that, in a preferred embodiment of the present invention, the average brightness variation amplitude threshold and the brightness variance increment threshold can be set based on statistical analysis of background noise distribution and typical pedestrian disturbance samples. Specifically, the average brightness variation amplitude threshold is used to distinguish between weak background disturbances and significant brightness changes. Preferably, it is set as the average brightness variation amplitude plus three times the standard deviation by statistically analyzing the brightness variation amplitude of the region under undisturbed conditions. For example, when the average brightness variation amplitude is 2.0 and the standard deviation is 1.2 in a undisturbed scene, the average brightness variation amplitude threshold can be set to 5.6.

[0127] The brightness variance increment threshold is used to measure the spatial non-uniformity of brightness distribution in the disturbed area. Preferably, the lower limit threshold that can distinguish between pedestrian occlusion and overall illumination fluctuation scenarios is selected by comparing samples. For example, when the average brightness variance increment is 18 when there is pedestrian disturbance and the average brightness variance increment is 6 when there is overall illumination fluctuation, the brightness variance increment threshold can be set to 12.

[0128] like Figure 4 As shown, the method for identifying Q traversable gap regions corresponding to a pedestrian's forward traversing trend includes:

[0129] S400: Let the initial value of m be 1, and the range of m is from 1 to M;

[0130] S401: Obtain the light and shadow disturbance feature set of the m-th traversable gap region, and extract the light and shadow expansion direction vector set and the disturbance centroid movement velocity set from the light and shadow disturbance feature set;

[0131] The average light and shadow expansion direction vector is calculated based on the set of light and shadow expansion direction vectors; the average perturbation centroid movement velocity is calculated based on the set of perturbation centroid movement velocities.

[0132] Convert the average light and shadow expansion direction vector into a light and shadow expansion direction angle;

[0133] Based on the active disturbance region of the current image frame corresponding to the m-th traversable gap region, extract the center coordinates of the corresponding active disturbance region.

[0134] Predict the future location of a target pedestrian per unit time based on the direction angle of light and shadow expansion and the center coordinates of the activity disturbance area;

[0135] If the target pedestrian's position exceeds the boundary of the m-th traversable gap area, it is considered that the target pedestrian has a tendency to rush into the lane, and the m-th traversable gap area is marked as a traversable gap area with a tendency for the pedestrian to cross forward.

[0136] S402: Let m = m + 1. If m is less than or equal to M, return to S401 and continue execution; if m is greater than M, the number of traversable gap areas marked as having a pedestrian forward crossing trend is recorded as Q, that is, Q traversable gap areas with a pedestrian forward crossing trend are obtained, and the current process ends.

[0137] The method for converting the direction angle of light and shadow expansion includes:

[0138]

[0139] Where, θ m Let be the direction angle of light and shadow expansion in the m-th passable gap region. Let be the average light and shadow propagation direction vector of the m-th traversable gap region, and arctan2(·) be the arctangent function.

[0140] The method for obtaining the target pedestrian location per unit time in the future includes:

[0141]

[0142] Among them, (X' m Y' m () represents the predicted future pedestrian location per unit time for the m-th traversable gap region. The coordinates of the center of the active disturbance region in the image frame corresponding to the m-th traversable gap region are given. Let ΔDWSJ be the average perturbation centroid velocity corresponding to the m-th traversable gap region, and θ be the future unit time. m The direction angle of light and shadow expansion corresponding to the m-th passable gap region.

[0143] It should be noted that, in a preferred embodiment of the present invention, by predicting the future location of a target pedestrian based on the light and shadow expansion direction vector and the velocity of the disturbance centroid, and determining the spatial relationship with the boundary of the traversable gap region, accurate identification of potential pedestrian crossing behavior can be achieved. Specifically, after obtaining the light and shadow disturbance feature set for each traversable gap region, the average light and shadow expansion direction vector and the average velocity of the disturbance centroid are calculated based on the disturbance dynamic information of multiple consecutive frames to obtain the overall motion trend of the light and shadow disturbance. The average light and shadow expansion direction vector is used to reflect the main expansion direction of the disturbance region contour in the time dimension. By performing an angle transformation on the average light and shadow expansion direction vector, the light and shadow expansion direction angle is obtained, which facilitates subsequent trajectory extrapolation and region relationship determination.

[0144] After determining the trend of the disturbance movement, the system further extracts the center coordinates of the disturbance region in the current frame. Using these center coordinates as the prediction starting point, and combining them with the light and shadow expansion direction angle and the average disturbance centroid movement velocity, the system predicts the position of the target disturbance in the image plane within a unit time interval. This predicted position can simulate the possible trajectory of the disturbance target in a short period of time, especially for determining whether it has the tendency to break through the gap boundary and move towards the lane.

[0145] By making an inclusion judgment between the predicted location and the boundary area of ​​the traversable gap, when the predicted location exceeds the boundary of the gap area, the disturbance area is considered to have a forward traversal tendency, and thus the gap area is marked as a potential pedestrian traversal risk area. Compared with strategies based solely on instantaneous velocity or positional relationships, this judgment method can more fully consider the directionality of the disturbance, velocity stability, and spatial spillover characteristics, significantly improving the accuracy and anticipation of identifying sudden pedestrian traversal trends in non-line-of-sight occlusion scenarios, and providing reliable prior target set support for subsequent hazard level assessment.

[0146] like Figure 5 As shown, the assessment method for the hazard level corresponding to the Q traversable gap regions includes:

[0147] S500: Let the initial value of q be 1, and the range of q is from 1 to Q;

[0148] S501: Obtain the center coordinates of the active disturbance region corresponding to the qth traversable gap region; calculate the predicted rendezvous distance based on the center coordinates of the active disturbance region and the current position of the vehicle;

[0149] The emergency braking distance is calculated based on the vehicle's current speed.

[0150] Add the emergency braking distance to the preset safety redundancy distance threshold one to obtain safety distance threshold one; add the emergency braking distance to the preset safety redundancy distance threshold two to obtain safety distance threshold two; safety distance threshold one is less than safety distance threshold two.

[0151] If the predicted rendezvous distance is greater than or equal to the second safe distance threshold, the hazard level is set to low hazard level; if the predicted rendezvous distance is greater than or equal to the first safe distance threshold and less than the second safe distance threshold, the hazard level is set to medium hazard level; if the predicted rendezvous distance is less than the first safe distance threshold, the hazard level is set to high hazard level.

[0152] S502: Let q = q + 1. If q is less than or equal to Q, return to S501 and continue execution; if q is greater than Q, obtain the danger level corresponding to Q traversable gap regions.

[0153] The method for calculating the rendezvous prediction distance includes:

[0154]

[0155] Among them, JHJL q Let X' be the predicted intersection distance between the vehicle's current position and the q-th traversable gap region. q Y' q () represents the center coordinates of the active disturbance region corresponding to the q-th traversable gap region. This indicates the vehicle's current location.

[0156] The method for calculating the emergency braking distance includes:

[0157]

[0158] ZDJL represents emergency braking distance, which refers to the distance the vehicle travels from the moment the driver begins braking until the vehicle comes to a complete stop. CS curr Let α be the vehicle's current speed, and α be the vehicle's average deceleration under the current road conditions; therefore, α is a negative number.

[0159] It should be noted that, in a preferred embodiment of this application, a hazard level assessment method based on a comparison of the predicted intersection distance and a safe distance threshold is used to achieve a quantitative classification and judgment of the potential pedestrian crossing risk. Specifically, the system extracts the center coordinates of the active disturbance area for each traversable gap area with a crossing trend, and calculates the predicted intersection distance by combining it with the current position of the driving vehicle. This distance is used as a quantitative indicator to measure the spatial proximity of the pedestrian disturbance target to the vehicle within the prediction time window.

[0160] Simultaneously, the system calculates the emergency braking distance based on the vehicle's current speed using a uniform deceleration motion model. Then, it superimposes two preset safety redundancy distance thresholds (one and two) to generate multi-level safety distance thresholds. These thresholds can be set according to road conditions, vehicle braking performance, and system safety strategies to construct a multi-layered dynamic safety buffer zone. For example, safety redundancy distance threshold one can be set to 5 meters, and safety redundancy distance threshold two can be set to 10 meters.

[0161] By comparing the predicted intersection distance with two levels of safety distance thresholds, the system can accurately perceive the pedestrian disturbance trend and, in conjunction with the vehicle's current driving status, dynamically distinguish the spatial proximity and collision potential of pedestrian disturbance targets. This enables precise identification and graded warning of different levels of danger, significantly improving the system's ability to perceive pedestrian crossing risks and its response timeliness under non-line-of-sight occlusion conditions.

[0162] The hazard warning module executes a hazard classification adaptive warning strategy based on the hazard levels of Q traversable gap areas.

[0163] like Figure 3 As shown, the method for implementing an adaptive early warning strategy based on the hazard level of Q traversable gap regions includes:

[0164] S600: Let the initial value of q be 1, and the range of q is from 1 to Q;

[0165] S601: Obtain the danger level corresponding to the qth passable gap region;

[0166] If the danger level is high, the system will immediately generate an emergency audible and visual alarm signal, and warn the driver through the dashboard warning lights, buzzer or vehicle head-up display device, indicating that there is a potential risk of pedestrians crossing the road; the system will calculate the vehicle's target deceleration speed based on the safe distance threshold corresponding to the danger level, and automatically trigger deceleration control to adjust the vehicle's current speed to the target deceleration speed.

[0167] If the danger level is moderate, the system generates a warning signal to prompt the driver to increase attention and prepare to slow down through visual or auditory means.

[0168] If the hazard level is low, the system will continue to monitor the area crossing the gap.

[0169] S602: Let q = q + 1. If q is less than or equal to Q, return to S601 and continue execution; if q is greater than Q, end the current process.

[0170] The method for calculating the target deceleration speed of the vehicle includes:

[0171]

[0172] Among them, CS aim Let YZ1 be the target deceleration speed of the vehicle, YZ1 be the first safety distance threshold, and RY1 be the first safety redundancy distance threshold corresponding to the first safety distance threshold. The target deceleration speed of the vehicle ensures that the current predicted rendezvous distance meets the safety distance threshold corresponding to the medium risk level.

[0173] It should be noted that, in a preferred embodiment of this application, a graded adaptive early warning strategy based on hazard level is used to dynamically adjust the early warning prompts and braking control responses according to the differences in hazard levels of traversable gap areas, thereby achieving intelligent tiered handling of sudden pedestrian crossing behavior. By dividing the hazard level into three levels—high risk, medium risk, and low risk—the system can implement differentiated control measures for different risk levels after identifying target areas with pedestrian crossing tendencies.

[0174] Specifically, when a high-risk level is detected, the system generates an emergency audible and visual alarm signal, calculates the vehicle's target deceleration speed based on a safe distance threshold, and automatically triggers braking control to reduce the vehicle speed to a target speed that meets medium-risk conditions. This proactive intervention in high-risk situations effectively reduces the probability of a potential collision. For medium-risk levels, the system provides visual or auditory cues to remind the driver to prepare for deceleration in advance, thereby enhancing the driver's sensitivity to potential risks. For low-risk levels, the system continuously monitors the disturbance state, dynamically updates the risk level, and adjusts the response strategy accordingly when the disturbance trend changes.

[0175] This application employs a tiered adaptive warning strategy, enabling the system to balance driving safety and the flexibility of warning intervention. Compared to traditional fixed thresholds or single braking strategies, it significantly improves the timeliness of identifying sudden pedestrian crossing risks, the accuracy of warning response, and the adaptability of braking intervention. This helps reduce false alarms and unnecessary braking operations, enhancing the overall safety protection capability and user experience of the intelligent driving system in complex road environments.

[0176] Example 2:

[0177] Please see Figure 2 As shown, this embodiment provides a method for early warning in intelligent driving, including:

[0178] Real-time detection of stationary vehicles on both sides of the road, and extraction of N traversable gap regions based on the spatial arrangement relationship and pixel spacing between the edges of adjacent vehicles;

[0179] Based on the image brightness change features extracted from N traversable gap regions, M traversable gap regions caused by the light and shadow disturbances caused by the activity of the target pedestrian are identified.

[0180] Based on the extraction of corresponding light and shadow disturbance feature sets within M traversable gap regions, Q traversable gap regions corresponding to pedestrians with a forward crossing trend are identified. Combined with the current position and speed of the vehicle, the corresponding danger level is assessed for the Q traversable gap regions.

[0181] Based on the hazard status levels of Q traversable gap regions, an adaptive early warning strategy for hazard status classification is implemented.

[0182] 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.

[0183] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent driving warning, characterized in that, The application relates to a method for real-time detection of a road, and more particularly to a method for real-time detection of a road and a method for real-time detection of a road. The method comprises the following steps: Real-time detection of static vehicles on both sides of the road, and extraction of N crossable gap regions based on the spatial arrangement relationship and pixel spacing between the edges of adjacent vehicles; Based on the image brightness variation features extracted from the N crossable gap regions, M crossable gap regions disturbed by light and shadow caused by target pedestrian activities are identified; Based on the corresponding light and shadow disturbance feature set extracted from the M crossable gap regions, Q crossable gap regions with a pedestrian crossing trend are identified, and the corresponding dangerous state levels of the Q crossable gap regions are evaluated in combination with the current position and speed of the vehicle. The method for obtaining the light and shadow disturbance feature set comprises the following steps: Divide the continuous images corresponding to each of the M crossable gap regions into a plurality of image frames, respectively extract the activity disturbance regions corresponding to adjacent image frames, perform edge contour extraction on each activity disturbance region, generate a minimum circumscribed rectangle and obtain the corresponding centroid coordinates; based on the centroid coordinates of adjacent image frames, a light and shadow expansion direction vector and a disturbance centroid moving speed are constructed, the light and shadow expansion direction vector is added to a light and shadow expansion direction vector set, and the disturbance centroid moving speed is added to a disturbance centroid moving speed set until all image frames are processed; the light and shadow expansion direction vector set and the disturbance centroid moving speed set are constructed into a corresponding light and shadow disturbance feature set; The method for identifying the Q crossable gap regions with a pedestrian crossing trend comprises the following steps: Based on the light and shadow disturbance feature set of each of the M crossable gap regions, the light and shadow expansion direction vector set and the disturbance centroid moving speed set are respectively extracted; Based on the light and shadow expansion direction vector set, an average light and shadow expansion direction vector is calculated, based on the disturbance centroid moving speed set, an average disturbance centroid moving speed is calculated, and the average light and shadow expansion direction vector is converted into a light and shadow expansion direction angle; Based on the center coordinates of the activity disturbance region of the current corresponding image frame of each crossable gap region, the target pedestrian position in the future unit time is predicted in combination with the light and shadow expansion direction angle and the center coordinates; When the predicted target pedestrian position exceeds the boundary of the corresponding crossable gap region, the crossable gap region is determined as a crossable gap region with a pedestrian crossing trend; finally, the Q crossable gap regions disturbed by light and shadow caused by target pedestrian activities are obtained; 2. The method for intelligent driving early warning according to claim 1, characterized in that, Based on the dangerous state levels of the Q crossable gap regions, a dangerous state grading adaptive warning strategy is executed. The method for executing the dangerous state grading adaptive warning strategy based on the dangerous state levels of the Q crossable gap regions comprises the following steps: The corresponding dangerous state levels of the Q crossable gap regions are obtained respectively; When the dangerous state level is a high-danger level, the system generates an emergency sound and light alarm signal, sends a warning to the driver through an instrument panel warning light, a buzzer or a vehicle head-up display device, calculates a vehicle target deceleration speed based on a safety distance threshold corresponding to the high-danger level, and automatically triggers deceleration control to adjust the current speed of the vehicle to the vehicle target deceleration speed. When the danger state level is a medium danger level, the system generates a pre-warning prompt signal to prompt the driver to pay attention and prepare for deceleration through vision or hearing; When the danger state level is a low danger level, the system keeps monitoring the crossable gap region continuously.

3. The method for intelligent driving early warning according to claim 1, characterized in that, The method for obtaining the active disturbance region comprises: Obtaining the average luminance variation amplitude and the luminance variance increment corresponding to the crossable gap region image frame; Marking the pixel points in the crossable gap region image frame as active disturbance pixel points when the average luminance variation amplitude is greater than a preset average luminance variation amplitude threshold and the luminance variance increment is greater than a preset luminance variance increment threshold; Constructing all the active disturbance pixel points into an active disturbance region.

4. The method for intelligent driving early warning according to claim 1, characterized in that, The method for evaluating the danger state level corresponding to the Q crossable gap regions comprises: For each crossable gap region with a forward crossing trend of the pedestrian, obtaining the center coordinates of the corresponding active disturbance region, calculating the intersection prediction distance based on the center coordinates of the active disturbance region and the current position of the vehicle, calculating the emergency braking distance based on the current speed of the vehicle, adding the emergency braking distance to a preset safety redundancy distance threshold one to obtain a safety distance threshold one, and adding the emergency braking distance to a preset safety redundancy distance threshold two to obtain a safety distance threshold two; When the intersection prediction distance is greater than or equal to the safety distance threshold two, the corresponding danger state level is set to a low danger level; when the intersection prediction distance is greater than or equal to the safety distance threshold one and less than the safety distance threshold two, the corresponding danger state level is set to a medium danger level; and when the intersection prediction distance is less than the safety distance threshold one, the corresponding danger state level is set to a high danger level.

5. The method for intelligent driving early warning according to claim 1, characterized in that, The method for extracting the image luminance variation features of the N crossable gap regions comprises: Dividing the continuously collected crossable gap region images into T image frames, calculating the average luminance variation amplitude and the luminance variance increment of adjacent tth frame and t-1th frame for each crossable gap region in sequence, dividing the image pixel gray scale into B equal-interval amplitude intervals, counting the number of pixel points in each amplitude interval and calculating the corresponding luminance disturbance entropy, adding the average luminance variation amplitude, the luminance variance increment and the luminance disturbance entropy to the corresponding set respectively, and constructing each set into the image luminance variation features corresponding to the crossable gap region after all the image frames are processed, until all the N crossable gap regions are processed, to obtain the image luminance variation features of the N crossable gap regions.

6. The method for intelligent driving early warning according to claim 1, characterized in that, The method for identifying the M crossable gap regions disturbed by the light and shadow caused by the target pedestrian activity comprises: Inputting the image luminance variation features corresponding to the N crossable gap regions into a pedestrian activity confidence evaluation model respectively to obtain the corresponding pedestrian activity confidence, and determining the crossable gap region corresponding to the pedestrian activity confidence greater than a preset pedestrian activity confidence threshold as the crossable gap region disturbed by the light and shadow caused by the target pedestrian activity, to obtain the M crossable gap regions disturbed by the light and shadow caused by the target pedestrian activity.

7. The method for intelligent driving early warning according to claim 1, characterized in that, The method for obtaining the N crossable gap regions comprises: Using a target detection algorithm to extract the static vehicle target from the input image and construct a static vehicle region. The edge positions of adjacent vehicles in the static vehicle area are sorted, and an edge arrangement sequence is constructed; For each pair of adjacent vehicles in the edge arrangement sequence, the pixel distance of the gap between the adjacent vehicles in the input image is calculated, and combined with the camera intrinsic matrix, the shooting height and the vehicle forward IMU attitude parameters of the current input image, the pixel distance is converted into the gap width value in the actual space using monocular projection geometry model; All the calculated gap width values are compared with the preset crossable width threshold, and the gap between the adjacent vehicles with a gap width value greater than the crossable width threshold is marked as a crossable gap region, that is, N crossable gap regions are obtained.

8. A system for intelligent driving warning, for implementing the method for intelligent driving warning according to any one of claims 1-7, characterized in that, Comprise: The gap region identification module is used for real-time detection of the static vehicles on both sides of the road, and the spatial arrangement relationship between the edges of adjacent vehicles and the pixel distance are used to extract N crossable gap regions; The activity feature diagnosis module identifies M crossable gap regions caused by light and shadow disturbance of the target pedestrian activity based on the image brightness change features extracted from the N crossable gap regions; The crossing risk assessment module identifies Q crossable gap regions corresponding to the pedestrian forward crossing trend based on the corresponding light and shadow disturbance feature set extracted from the M crossable gap regions, and evaluates the corresponding dangerous state level for the Q crossable gap regions in combination with the current position and the current speed of the vehicle; The dangerous state warning module executes the dangerous state classification adaptive warning strategy based on the dangerous state level of the Q crossable gap regions.

Images