CN-121982904-A - Vehicle three-dimensional overrun intelligent detection method and system

CN121982904ACN 121982904 ACN121982904 ACN 121982904ACN-121982904-A

Abstract

The application belongs to the technical field of traffic monitoring and discloses a vehicle three-dimensional overrun intelligent detection method and system, wherein the method comprises the steps of acquiring image information of a road area where a target vehicle is positioned, and identifying a static road reference object with standard physical size in the image information; the method comprises the steps of constructing a scale field model of a road area based on pixel sizes of static road references in image information and standard physical sizes corresponding to the pixel sizes, extracting feature point coordinates of a target vehicle in the image information, determining a direction angle of the target vehicle, calculating the three-dimensional physical size of the target vehicle according to the scale field model, the feature point coordinates and the direction angle, obtaining statistical size parameters corresponding to vehicle types of the target vehicle, comparing the three-dimensional physical size with the statistical size parameters to determine an overrun state of the target vehicle, and therefore detection accuracy and reliability can be improved, environment adaptability can be improved, misjudgment rate is reduced, and a calibration object is not required to be preset.

Inventors

WU BINGXIN
ZENG CHUNCHAO
CHEN HONGJUN
HUO QIFENG

Assignees

广东省科学院佛山产业技术研究院有限公司

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (10)

1. The three-dimensional overrun intelligent detection method for the vehicle is characterized by comprising the following steps of: A1. Acquiring image information of a road area where a target vehicle is located, and identifying a static road reference object with standard physical dimensions in the image information; A2. Constructing a scale field model of the road area based on the pixel size of the static road reference object in the image information and the standard physical size corresponding to the pixel size, wherein the scale field model reflects the mapping relation between the pixel size and the physical size of each pixel of the road surface of the road area; A3. Extracting feature point coordinates of a target vehicle in the image information, and determining a direction angle of the target vehicle; A4. calculating the three-dimensional physical dimension of the target vehicle according to the scale field model, the characteristic point coordinates and the direction angle, wherein the three-dimensional physical dimension comprises height, length and width; A5. and acquiring a statistical size parameter corresponding to the vehicle type of the target vehicle, and comparing the three-dimensional physical size with the statistical size parameter to determine the overrun state of the target vehicle.
2. The three-dimensional overrun intelligent detection method of a vehicle according to claim 1, further comprising the steps of, after step A5: A6. And carrying out hierarchical alarm operation according to the overrun state of the target vehicle.
3. The vehicle three-dimensional overrun intelligent detection method of claim 1, wherein the static road references comprise at least one of traffic markings, traffic signs, and fixed assets.
4. The three-dimensional overrun intelligent detection method of a vehicle according to claim 1, wherein the step A2 includes: A201. Acquiring an image sequence formed by continuous multi-frame image information; A202. for each observation point on the static road reference object, acquiring the standard physical size corresponding to the observation point and the pixel size of the corresponding static road reference object in the image sequence of the observation point, and calculating a scale observation value sequence corresponding to the image sequence of the observation point, wherein the scale observation value represents the pixel size corresponding to the unit physical length at the observation point; A203. Correcting the scale observation value sequence, and acquiring an effective scale observation value according to the corrected scale observation value sequence, wherein the correcting process comprises time domain filtering process, outlier process and missing value supplementing process; A204. and constructing a scale field model of the road area by using a Gaussian process regression method according to the effective scale observed value of each observation point.
5. The method for three-dimensional overrun intelligent detection of a vehicle according to claim 4, wherein step a204 comprises: Acquiring pixel coordinates and quality parameters of each observation point, wherein the quality parameters comprise confidence, integrity and type of a static road reference object corresponding to the observation point; according to the quality parameters, calculating the fusion weight of each observation point; And constructing a scale field model of the road area by using a Gaussian process regression method according to the effective scale observation value, the pixel coordinates and the fusion weight of each observation point, wherein the Gaussian process regression method adopts a Matrn 3/2 kernel function as a covariance function, and inversely mapping the fusion weight into an observation noise variance in Gaussian process regression.
6. The three-dimensional overrun intelligent detection method of a vehicle according to claim 1, wherein the step A3 includes: A301. extracting pixel coordinates of feature points of a target vehicle in current frame image information as the feature point coordinates, wherein the feature points comprise a front-most point of a vehicle head, a rear-most point of the vehicle tail, a left-side outermost point, a right-side outermost point and a vehicle roof highest point; A302. Detecting contact points of front and rear wheels of a target vehicle and a road surface in current frame image information, mapping the contact points to a road surface physical coordinate system based on the scale field model, and calculating a direction angle of the target vehicle according to the mapped contact points to obtain a first direction angle estimated value; A303. Tracking the contact point of the front wheel or the rear wheel of the target vehicle with the road surface in continuous multi-frame image information to obtain a track point sequence under an image coordinate system, mapping the track point sequence to a road surface physical coordinate system based on a scale field model, and calculating the direction angle of the target vehicle according to the mapped track point sequence to obtain a second direction angle estimated value; A304. And fusing the first direction angle estimated value and the second direction angle estimated value to obtain a final direction angle.
7. The method for three-dimensional overrun intelligent detection of a vehicle according to claim 6, wherein step A4 comprises: A401. Constructing a pavement equation according to the scale field model; A402. calculating the height of the target vehicle through a geometric relationship by utilizing the pixel coordinates of the highest point of the vehicle roof and the contact point of the front wheel and the rear wheel of the target vehicle with the road surface and combining the camera parameters and the road surface equation; A403. Determining scale values of the front-most point, the rear-most point, the left-most point and the right-most point of the vehicle head based on pixel point coordinates of the front-most point, the rear-most point, the left-most point and the right-most point of the vehicle head and the scale field model; A404. Mapping the corresponding characteristic points to a road surface physical coordinate system based on scale values of a front-most point, a rear-most point, a left-most outer point and a right-most outer point of the vehicle head to obtain three-dimensional coordinates of the front-most point, the rear-most outer point, the left-most outer point and the right-most outer point of the vehicle head; A405. according to the three-dimensional coordinates of the front-most point and the rear-most point of the vehicle head and the final direction angle, calculating the projection distance of the front-most point and the rear-most point of the vehicle head in the longitudinal axis direction of the target vehicle to obtain the length of the target vehicle; A406. and calculating the projection distance of the left outermost point and the right outermost point in the transverse axis direction of the target vehicle according to the three-dimensional coordinates of the left outermost point and the right outermost point and combining the final direction angle to obtain the width of the target vehicle.
8. The three-dimensional overrun intelligent detection method of a vehicle according to claim 1, wherein the step A5 includes: A501. Identifying a vehicle model of the target vehicle; A502. inquiring in a vehicle size statistical database according to the vehicle type to obtain corresponding statistical size parameters, wherein the statistical size parameters comprise the mean value and standard deviation of the height, the length and the width; A503. determining effective confidence intervals of the height, the length and the width according to the average value and the standard deviation of the queried height, length and width; A504. And comparing the height, length and width of the target vehicle with the corresponding effective confidence intervals, and judging the overrun state of the target vehicle.
9. The method for three-dimensional overrun intelligent detection of a vehicle according to claim 8, further comprising the steps of, after step a 504: A505. If the target vehicle is not out of limit, updating the statistical size parameters of the corresponding vehicle type in the vehicle type size statistical database according to the three-dimensional physical size of the target vehicle.
10. The three-dimensional overrun intelligent detection system for the vehicle is characterized by comprising a camera and a control system; the camera is used for collecting image information of a road area where the target vehicle is located; The control system is used for identifying a static road reference object with standard physical dimensions in the image information, constructing a scale field model of a road area based on the pixel dimensions of the static road reference object in the image information and the standard physical dimensions corresponding to the pixel dimensions, wherein the scale field model reflects the mapping relation between the pixel dimensions and the physical dimensions of each pixel of a road surface of the road area, extracting characteristic point coordinates of a target vehicle in the image information, determining the direction angle of the target vehicle, calculating the three-dimensional physical dimensions of the target vehicle according to the scale field model, the characteristic point coordinates and the direction angle, wherein the three-dimensional physical dimensions comprise height, length and width, acquiring statistical dimension parameters corresponding to the vehicle type of the target vehicle, and comparing the three-dimensional physical dimensions with the statistical dimension parameters to determine the overrun state of the target vehicle.

Description

Vehicle three-dimensional overrun intelligent detection method and system Technical Field The application relates to the technical field of traffic monitoring, in particular to a three-dimensional overrun intelligent detection method and system for a vehicle. Background Overrun transportation is a serious problem of damaging road safety and infrastructure, and three-dimensional overrun refers to overrun conditions of three dimensions of length, width and height. The existing three-dimensional overrun detection technology mainly has the following technical bottlenecks: Although the video measurement technology based on fixed calibration can realize a certain degree of size measurement, the method has obvious limitations. The camera calibration is carried out by relying on calibration patterns specially arranged on the road surface of the monitored area, and the preset calibration objects are easily influenced by pollution, shielding and environmental changes. Even more serious, once the camera is displaced or the angle is changed, calibration needs to be carried out again, so that the maintenance cost of the system is high. Although the video measurement technology based on the deep learning regression avoids the requirement of preset calibration objects, the technical defects of the technology are also prominent. The method is seriously dependent on the quality and quantity of training data, and shows poor generalization capability in practical application. Due to the "black box" nature of the deep learning model, the decision process lacks interpretability, which makes it difficult for the measurement results to provide convincing evidence support in law enforcement. Although the method based on a single natural reference is attempted to use the existing road elements such as lane line width as a reference, the method is prone to failure when the reference is missing, blurred or occluded. More importantly, the method cannot fully consider the change characteristic of perspective distortion in space, so that the measurement accuracy is difficult to meet the actual requirement. The prior art has the following core defects that the detection dimension is not comprehensive, particularly the detection precision and reliability of ultra-long and ultra-wide vehicles are poor, the environment adaptability is weak, the influence of illumination change, bad weather and shielding is easy, the system vulnerability is high, the standards are excessively dependent on preset, fixed and single measurement standards, the standards are extremely easy to fail in an actual road environment, the misjudgment rate is high, the intelligent cognition on the relevance of the type and the size of the vehicles is lacking, the large-scale vehicles with compliance are frequently misreported as overrun, the functions are single, most systems can only provide simple yes/no judgment, cannot output detailed reports with quantitative evidence, the deployment flexibility is poor, the requirements on the installation parameters of cameras are strict, and the spherical cameras with variable visual angles are difficult to adapt. Of particular concern, the prior art generally lacks effective mechanisms for maintaining measurement stability under complex interference conditions, such as dynamic occlusion, severe light changes, and severe weather scenarios. Meanwhile, the technologies cannot fully utilize the existing diversified marking resources of the road to carry out intelligent complementary measurement, so that the reliability and the adaptability of the system in practical application are obviously insufficient. In view of the above, there is a need in the art for improvements. Disclosure of Invention The application aims to provide a three-dimensional overrun intelligent detection method and system for a vehicle, which can improve detection precision and reliability, enhance environment adaptability, reduce misjudgment rate and avoid presetting a calibration object. In a first aspect, the application provides a vehicle three-dimensional overrun intelligent detection method, which comprises the following steps: A1. Acquiring image information of a road area where a target vehicle is located, and identifying a static road reference object with standard physical dimensions in the image information; A2. Constructing a scale field model of the road area based on the pixel size of the static road reference object in the image information and the standard physical size corresponding to the pixel size, wherein the scale field model reflects the mapping relation between the pixel size and the physical size of each pixel of the road surface of the road area; A3. Extracting feature point coordinates of a target vehicle in the image information, and determining a direction angle of the target vehicle; A4. calculating the three-dimensional physical dimension of the target vehicle according to the scale field model, the characteristic point coordinates