A low-complexity unmanned aerial vehicle-based road anomaly recognition method

Abstract

The application discloses a low-complexity unmanned aerial base road anomaly identification method and belongs to the field of computer vision; specifically, firstly, road images with parallel line structures are collected according to continuous time sequences, edge features of each frame of image are extracted, and then edge straight lines are extracted, and the intersection point coordinate median of all non-collinear straight lines is the vanishing point coordinate; then, the estimated attitude angle of the unmanned aerial vehicle is derived, compared with the attitude angle measured by the sensor to eliminate the error of unmanned aerial vehicle jitter, and then the attitude angle deviation is calculated; the detection time is segmented, and the road detection unit is corresponded; elements with the attitude angle deviation exceeding a threshold value are marked as abnormal; a road detection unit with the number of abnormal elements exceeding a threshold value is found out, and the difference between each pair of adjacent attitude angle deviations and the time difference of the unit are calculated; when the deviation mean or the time difference mean is greater than or equal to a threshold value, the road detection unit has an anomaly, an alarm is given, and output is performed. The application saves the time of the unmanned aerial vehicle for executing a road anomaly detection task.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of unmanned system application and computer vision technology, and specifically relates to a low-complexity unmanned air-based road anomaly identification method. BACKGROUND

[0002] Roads are important tools for ensuring modern transportation; however, after accidents or natural disasters, road collapses and blockages caused by foreign objects may cause road traffic to be interrupted, which causes great obstacles to goods transportation, disaster relief and other work.

[0003] In order to improve the efficiency and quality of transportation work, it is necessary to timely grasp the information such as the opening and closing of the road and the damage degree, which is an important basis for guiding the subsequent work. The unmanned aerial vehicle can better adapt to the application scene of road anomaly detection due to its flexible maneuvering, all-weather task execution and high autonomy.

[0004] In the aspect of unmanned aerial vehicle road anomaly detection, the existing work mostly focuses on the application of pattern recognition. First, the prior knowledge picture of the road anomaly area is trained, and after the unmanned aerial vehicle completes the aerial photography, the picture obtained by the aerial photography is directly tested and classified to screen out the picture with road anomaly, and then the anomaly is marked. This method has a high recognition rate. However, when the edge features of the road picture are extracted, the whole picture (containing various environmental information) needs to be processed, so that the detection algorithm has high complexity. At the same time, due to the complex causes and various situations of road anomalies, it is difficult to enumerate all road anomaly situations, so it is difficult to meet the requirements of prior knowledge for pattern recognition in actual situations.

[0005] Generally speaking, the two edges of the road have the typical parallel line feature. According to the perspective principle of geometric optics, the parallel line in the three-dimensional world will intersect at a point when projected onto the two-dimensional plane. The intersection point is the vanishing point. Whether the vanishing point exists deviation reflects whether the feature of the road parallel line changes, and the vanishing point deviation can be reflected by the camera coordinate system pose estimated by the image coordinate of the vanishing point. That is, in the image coordinate system, the pose information (yaw angle and pitch angle) of the camera carried by the unmanned aerial vehicle can be estimated through coordinate transformation according to the position coordinates of the vanishing point in the image at each time. If the measured camera pose at this time does not coincide with the measured camera pose at the previous time, it can be explained that the vanishing point in the image coordinate has a deviation.

[0006] In actual situation, if the road appears vehicle congestion, obstacle blockage and road fracture, etc., the road parallel line feature will be destroyed, so that the vanishing point detected by the unmanned aerial vehicle detection system repeatedly deviates around a region in the two-dimensional image in a period of time, that is, the vanishing point jitter. The jitter degree of the vanishing point reflects whether the parallel line feature of the road changes obviously, if it changes obviously, it is considered that the road exists abnormality. SUMMARY

[0007] In order to solve the problems of high complexity of existing pattern recognition algorithm and lack of prior knowledge of road abnormality when the unmanned aerial vehicle performs road abnormality detection task, the application provides a low complexity unmanned aerial base road abnormality recognition method, which combines computer vision high recognition rate, detects the jitter degree of the vanishing point to determine whether the parallel line feature of the road changes, and then determines whether the road exists abnormality, thereby reducing the algorithm complexity.

[0008] The low complexity unmanned aerial base road abnormality recognition method has the following specific steps:

[0009] Step one, for the road with parallel line structure, the road image is collected according to the continuous time sequence;

[0010] Step two, the edge feature of each frame image is extracted, the edge straight line of the processed target road area is extracted to obtain a plurality of straight lines, the median value of the intersection point coordinates of all non-collinear straight lines is calculated to obtain the vanishing point o of the target road area in the frame image and the coordinates of the vanishing point o in the image coordinate system

[0011] Step three, the vanishing point coordinates of the nth frame image are combined with the intrinsic matrix of the camera to derive the estimated pitch angle and the estimated yaw angle of the unmanned aerial vehicle at the current time, and the two are defined as the estimated attitude angle of the unmanned aerial vehicle;

[0012] Firstly, the vanishing point vector of the nth frame image is defined according to the vanishing point coordinates

[0013] Then, the third column of the rotation matrix R is calculated according to the intrinsic matrix I of the camera calibration n

[0014] Finally, the third column of the rotation matrix is substituted into to obtain the pitch angle α n and the yaw angle β n of the unmanned aerial vehicle at the time corresponding to the nth frame image;

[0015]

[0016] The calculation formula is:

[0017] ​

[0018]

[0019] For the expression of the n-th frame image contains the pitch angle and yaw angle information of the intermediate vector.

[0020] Step four, the current time the n-th frame image derived from the estimated attitude angle, and the actual measurement of the UAV attitude angle, to determine whether to meet If yes, update the estimated attitude angle is And calculate the road target area in the n-th frame image of the vanishing point jitter corresponding to the UAV attitude angle deviation Δθ n ; Otherwise, directly calculate Δθ n ;

[0021] The current time the estimated attitude angle of the UAV is

[0022] The current time the actual measurement of the UAV attitude angle is

[0023] Threshold, used to determine whether the measurement of the UAV attitude angle and the estimated value has a significant deviation, too large deviation is considered to be the cause of the vanishing point jitter is caused by the UAV jitter.

[0024] UAV attitude angle deviation Δθ n The formula is:

[0025]

[0026] Step five, the overall collection time is divided into intervals I time period, each time period is divided into intervals K time period, each continuous K attitude angle deviation Δθ n And the corresponding time to form a deviation-time set, each deviation-time set corresponds to a road detection unit;

[0027] The K-th attitude angle deviation corresponding to the i-th road detection unit constitutes a deviation-time set U i :

[0028]

[0029] Step six, in each road detection unit, to determine whether the corresponding UAV attitude angle deviation Δθ i The modulus value of the image is greater than the set distance threshold If yes, mark the image corresponding to the deviation as abnormal; otherwise, do not process the image;

[0030] Step seven, record the number of images marked as abnormal in each deviation-time set, and determine whether the number of abnormalities is less than the set threshold N. If so, it is considered that the road detection unit has no abnormality, and the next road detection unit is continued to be judged; otherwise, step eight is entered;

[0031] Step eight, in the road detection unit where the number of abnormal images exceeds the threshold N, all abnormal images are selected, the difference value of the attitude angle deviation of each pair of adjacent abnormal images is calculated and the time difference Δt corresponding to the moment, and then the mean value of the attitude angle deviation difference and the time difference is calculated E(Δt);

[0032] Step nine, continue to judge whether the mean values of the deviation difference and the time difference E(Δt) are less than the respective thresholds and T Δt , if so, it is considered that the road detection unit has no abnormality; otherwise, when any mean value is greater than or equal to the corresponding threshold, it is considered that the road detection unit has an abnormality, an alarm is given and the serial number corresponding to the abnormal detection unit and all frame images contained therein are output.

[0033] and T Δt respectively represent the values of the deviation of the attitude angle of the unmanned aerial vehicle and the value of E(Δt) when reaching the critical stability at the corresponding moment.

[0034] The advantages of the present application are:

[0035] 1) The low-complexity unmanned aerial base road anomaly recognition method greatly reduces the workload of the unmanned aerial vehicle road anomaly detection system and the requirement for data storage capacity compared with the traditional anomaly detection method, and saves the time of the unmanned aerial vehicle for executing the road anomaly detection task.

[0036] 2) The low-complexity unmanned aerial base road anomaly recognition method defines the road anomaly as: due to the uncontrollable factors, the normal road line parallel line structure is destroyed, thereby causing the objective road scene of traffic obstruction; the existence of road anomaly is detected through the shaking of the vanishing point, and the detection process does not depend on human participation, thereby providing a new idea for road anomaly detection.

[0037] 3) The low-complexity unmanned aerial base road anomaly recognition method compares the actual measured attitude angle of the unmanned aerial vehicle with the estimated attitude angle of the unmanned aerial vehicle through the sensor, and excludes the shaking of the vanishing point caused by the shaking of the unmanned aerial vehicle itself; the threshold value discrimination method is used to determine the concentration degree of the attitude angle anomaly in the angle deviation and the time, and the shaking of the vanishing point caused by the turning of the road and the rapid passing of foreign matters through the camera lens is excluded. BRIEF DESCRIPTION OF DRAWINGS​

[0038] Figure 1 is a low-complexity unmanned aerial road anomaly recognition method flowchart of the present application;

[0039] Figure 2 is a flowchart of the single-frame image vanishing point detection method of the present application;

[0040] Figure 3 is a flowchart of the method for reducing the false alarm rate of road anomaly detection. DETAILED DESCRIPTION

[0041] The specific embodiments of the present application will be further described in detail below with reference to the accompanying drawings and examples. The following examples or drawings are used to illustrate the present application, but not to limit the scope of the present application.

[0042] The present application gives an unmanned aerial road anomaly detection method by combining the high recognition rate of computer vision and the low complexity of the algorithm for detecting road anomalies using vanishing point jitter, with the constraint condition of detecting the abnormal situation that the parallel line structure of the road changes significantly.

[0043] The new road anomaly involved in the present application is defined as: due to uncontrollable factors, the parallel line structure of the normal road is destroyed, resulting in an objective road scene that causes traffic congestion.

[0044] The low-complexity unmanned aerial road anomaly recognition method described in the present application is based on the following devices:

[0045] An unmanned aerial vehicle: as a mobile platform for performing road anomaly recognition tasks.

[0046] A central processing unit: as the core unit for processing road anomaly detection algorithms and sensor data.

[0047] A camera (calibrated): needs to be rigidly connected to the unmanned aerial vehicle body and keep the lens and the nose consistent in orientation, responsible for shooting road scenes and processing into pictures as input for the road anomaly detection system.

[0048] An angle sensor: responsible for measuring the attitude angle information (yaw angle and pitch angle) of the unmanned aerial vehicle (or camera), with the pitch angle being 0° when the body is flying horizontally; the yaw angle is 0° when the vanishing point is located on the longitudinal axis of the image coordinate system.

[0049] The present application mainly involves the conversion between three coordinate systems:

[0050] 1) Camera coordinate system, with its origin at the optical center of the camera lens, the z-axis in the direction of the optical axis, and the x and y directions determined by the right-hand rule. This method only involves the pitch angle and yaw angle of the camera lens, so it is represented by (α, β);

[0051] 2) Pixel coordinate system, usually taking the upper left corner of the imaging plane as the origin, taking the 0th row and the 0th column as the two coordinate axes, the coordinate system is a two-dimensional coordinate system, commonly represented by (u, v);

[0052] 3) Image coordinate system, the origin of the coordinate system is at the intersection of the optical axis and the imaging plane, and the x and y axes are consistent with the directions of the u and v axes in the pixel coordinate system, which is represented by (x, y).

[0053] The method first acquires road images in a continuous time sequence for a road with parallel line structure, extracts edge features for each frame of image, extracts a plurality of straight lines from the edge straight line extraction of the processed target road region image, calculates the median of the intersection point coordinates of all non-collinear straight lines, and obtains the vanishing point coordinates of the target road region on the image. Then, the estimated attitude angle of the unmanned aerial vehicle at the current time is derived, compared with the actual attitude angle measured by the sensor of the unmanned aerial vehicle to eliminate the influence of the shaking of the unmanned aerial vehicle on the estimated attitude angle, and the attitude angle deviation corresponding to the shaking of the vanishing point is calculated. By segmenting the detection time, the road detection unit is corresponded; images whose attitude angle deviation exceeds the threshold value are marked as abnormal; the road detection unit whose number of abnormal images exceeds the threshold value is found, and the attitude angle deviation difference and the time difference of each pair of adjacent abnormal images are calculated. When the mean value of the deviation or the time difference is greater than or equal to the threshold value, the road detection unit is abnormal, an alarm is given and output.

[0054] The low-complexity unmanned aerial base road anomaly recognition method, as shown in Figure 1 , the specific steps are as follows:

[0055] Step 1: Mount a camera on the unmanned aerial vehicle, and acquire road images in a continuous time sequence for a road with parallel line structure;

[0056] First, the device is combined, and the angle sensor is installed to the unmanned aerial vehicle body; the camera is installed to the head or the bottom of the body, and the lens keeps consistent with the heading of the head.

[0057] When the unmanned aerial vehicle is flying, the relative height between the unmanned aerial vehicle and the ground should be kept at a fixed value;

[0058] Step 2: Perform semantic segmentation on each frame of road picture to obtain a segmentation mask picture of the road;

[0059] The semantic segmentation model is, for example, a BiSeNet v2 model. The reason why the mask picture processed by semantic segmentation is used as the input picture for detection is that it can shield many background interferences and reduce the missed detection and false detection.

[0060] Step three, the mask image is smoothed by Gaussian filter and median filter, the mask image gradient is calculated (such as using Canny operator), the gradient direction of each pixel is thresholded to form a binary image of encoded gradient direction;

[0061] Step four, the straight line features of the image are extracted by Hough transform to obtain the detected straight line, the intersection median of all non-collinear straight lines is calculated to obtain the vanishing point o of the frame image and the coordinates of the vanishing point o in the image coordinate system

[0062] The two edge lines of the road have typical parallel line features, according to the perspective principle of geometric optics, the parallel lines intersect at a point when projected onto a two-dimensional plane in the three-dimensional world, and the intersection point is the vanishing point.

[0063] The single-frame image vanishing point detection algorithm process is shown in the following table: Figure 2 In this embodiment, the operation process is briefly described as follows: first, the picture is smoothed by Gaussian filter and median filter (which is beneficial to remove salt and pepper noise);

[0064] Then, the picture gradient is calculated by using Canny operator; the threshold is set; after calculating the gradient direction of each pixel and thresholding, a binary image of encoded gradient direction can be formed, and then the Hough transform is used to extract the straight line features. After extracting the features, threshold judgment is needed. The threshold is used to set the slope range of the dominant straight line, to exclude some target straight lines that are not likely to appear in some slope range, and to reduce the calculation amount of the algorithm. The transformed points are voted in the Hough space; the voting points are inversely transformed to obtain the detected straight line, and then the image vanishing point and the image coordinates of the image vanishing point are obtained.

[0065] Step five, the estimated pitch angle and the estimated yaw angle of the unmanned aerial vehicle at the current time are derived from the vanishing point coordinates of the nth frame image and the intrinsic matrix of the camera;

[0066] First, the vanishing point vector of the nth frame image is defined according to the vanishing point coordinates

[0067] Then, the rotation matrix R is calculated according to the intrinsic matrix I of camera calibration n The third column of the rotation matrix is:

[0068]

[0069] The third column of the rotation matrix is:

[0070]

[0071] Finally, the third column of the rotation matrix is substituted into to obtain the pitch angle a of the unmanned aerial vehicle at the time corresponding to the nth frame imagen and yaw angle β n ;

[0072]

[0073]

[0074] is the intermediate vector for expressing the n-th frame image containing the pitch angle and yaw angle information.

[0075] The pitch angle and yaw angle of the UAV at the time corresponding to the n-th frame image are derived through the image coordinates of the vanishing point in the n-th frame image.

[0076] As shown in Figure 3 , steps six to eleven are the flow of the method for reducing the false alarm rate of road anomaly detection.

[0077] Step six, compare the estimated attitude angle derived from the n-th frame image at the current time with the actual measured attitude angle of the UAV, and determine whether it meets If yes, update the estimated attitude angle to and calculate the UAV attitude angle deviation Δθ n corresponding to the vanishing point jitter of the road target region in the n-th frame image; otherwise, directly calculate Δθ n .

[0078] The estimated attitude angle of the UAV at the current time is

[0079] The actual measured attitude angle of the UAV at the current time is

[0080] is a threshold value for determining whether there is a significant deviation between the measured value and the estimated value of the UAV attitude angle. If the deviation is too large, it is considered that the reason for the vanishing point jitter at this time is the UAV jitter.

[0081] The UAV attitude angle deviation Δθ n The calculation formula is:

[0082]

[0083] Step seven, divide the overall collection time into intervals I time periods, and then divide each time period into intervals K time periods. Form a deviation-time set with each continuous K attitude angle deviations Δθ n and the corresponding time, and each deviation-time set corresponds to a road detection unit.

[0084] Suppose that M frames of photos are obtained in the whole image acquisition process, and the overall acquisition time is T. The overall acquisition time T is equally divided into I time periods, and each time period contains K frames of images (corresponding to K attitude angle deviations), which satisfy I x K = M, i.e., T / I is the time period corresponding to the road detection unit, and K is the number of image frames (attitude angle deviations) in the time period.

[0085] The K attitude angle deviations corresponding to the ith road detection unit form a deviation-time set U i .

[0086]

[0087] Step eight, in each road detection unit, it is judged whether the modulus of each UAV attitude angle deviation Δθ i is greater than or equal to a set threshold value , i.e., the distance value is greater than or equal to If yes, the image element corresponding to the deviation is marked as abnormal; otherwise, the image is not processed.

[0088] Step nine, the number of images marked as abnormal in each deviation-time set is recorded, and it is judged whether the number of abnormalities is less than a set threshold value N. If yes, it is considered that the road detection unit is normal, and the next road detection unit is judged; otherwise, step ten is entered.

[0089] Step ten, in the road detection unit where the number of abnormal images exceeds the threshold value N, all abnormal images are selected, the difference value of the attitude angle deviation of each pair of adjacent abnormal images is calculated , and the time difference Δt corresponding to the time is calculated, and then the mean value of the deviation difference and the time difference is calculated E(Δt) as a statistical quantity for measuring the concentration degree of abnormal elements in the angle deviation and time.

[0090] Suppose that after deleting the deviation-time data corresponding to the normal elements in the road detection unit, there are p abnormal elements corresponding to the deviation-time data, the time difference of each pair of adjacent abnormal elements is calculated, and p-1 deviation difference and time difference sets (Δt1, Δt2,..., Δt p-1 ) are obtained, and the mean value of these distance differences is calculated:

[0091]

[0092] The mean value of the time difference is:

[0093]

[0094] ​(5) The formula can be used as a statistical quantity for measuring the concentration degree of abnormal elements in distance.

[0095] Step eleven, continue to judge the mean value of deviation difference and time difference E(Δt) whether respectively less than the respective threshold value And T Δt If yes, it is considered that the road detection unit is abnormal; otherwise, when any mean value is greater than or equal to the corresponding threshold value, it is considered that the road detection unit is abnormal, an alarm is given, and the serial number corresponding to the abnormal detection unit and all frame images contained therein are output.

[0096] And T Δt Respectively represent the deviation of the attitude angle of the unmanned aerial vehicle and the time when the critical stability is reached The value of E(Δt);

[0097] This step focuses on the number of abnormal elements greater than or equal to the threshold value The road detection unit, it should be noted that at this time, the road detection unit cannot be simply determined as abnormal. Because if the unmanned aerial vehicle is flying through an area with high road curvature during the time when this detection unit is located, it may also cause the number of abnormal elements to exceed the threshold value. Therefore, according to the characteristics that the image abnormal elements representing the area with high road curvature are more evenly distributed in deviation and time, and the characteristics that the image abnormal elements representing the real abnormal area of the road are not evenly distributed in distance or time, the road "false" abnormality and the real abnormality can be distinguished.

[0098] Further, the semantic segmentation model described in the present application is not limited to the BiSeNet v2 model, and any semantic segmentation model that can identify roads is suitable for the unmanned aerial vehicle-based road line anomaly detection method described in the present application. Similarly, the calculation of image gradients is not limited to using the Canny operator, and any operator that can obtain image gradients is suitable for the unmanned aerial vehicle-based road line anomaly detection method described in the present application.

[0099] The vanishing point detection method of the present application is currently mainly divided into three categories, namely using Hough transform to detect vanishing points, using RANSAC method to detect vanishing points, and a method involving an exhaustive search process for unknown entities to detect vanishing points. The above three methods have advantages and disadvantages in different scenarios, and the present application uses but is not limited to using Hough transform method to realize the vanishing point detection of road mask graph.

[0100] The road anomaly detection method emphasizes but is not limited to the anomaly detection of general roads. Any traffic path (such as a tunnel, a railway, a river embankment, a road, a pipeline, etc.) that meets the parallel line feature is suitable for the low-complexity unmanned air-based road anomaly recognition method.

Claims

1. A low complexity unmanned aerial road anomaly identification method, characterized in that, The specific steps are as follows: First, for the road with parallel line structure features, the road image is collected in continuous time sequence, the edge feature of each frame image is extracted, the edge straight line of the processed target road region image is extracted to obtain several straight lines, the median value of the intersection point coordinates of all non-collinear straight lines is calculated to obtain the vanishing point o of the target road region in the frame image and the coordinates of the vanishing point o in the image coordinate system coordinates of the vanishing point projected onto the frame image coordinate system for the road target region of the n-th image; Then, the vanishing point vector of the nth frame image is defined according to the vanishing point coordinates And the rotation matrix R is calculated according to the intrinsic parameter matrix I of camera calibration n The third column of the rotation matrix is substituted into, and the pitch angle a of the unmanned aerial vehicle at the moment corresponding to the nth frame image is estimated The third column of the rotation matrix is substituted into, and the pitch angle a of the unmanned aerial vehicle at the moment corresponding to the nth frame image is estimated n And the yaw angle β n ; The calculation formula is: to express the intermediate vector for the nth frame image includes pitch and yaw angle information; Then, according to the estimated attitude angle of the unmanned aerial vehicle at the moment corresponding to the n-th frame of image The actual attitude angle measured by the sensor of the unmanned aerial vehicle Subtract and take the modulus value to determine whether it meets is a threshold value; If yes, update the estimated attitude angle as And calculate the deviation of the UAV attitude angle Δθ corresponding to the vanishing point jitter of the road target region in the nth frame of image n ; otherwise, directly calculate Δθ n ; The whole collection time is divided into intervals I time periods, each time period is further divided into intervals K time periods, each continuous K attitude angle deviation Δθ n and corresponding time constitute deviation-time set, each deviation-time set corresponds to a road detection unit respectively; in each road detection unit, it is judged whether the modulus of the unmanned aerial vehicle attitude angle deviation Δθ i corresponding to each image exceeds the set threshold value If yes, mark the image corresponding to the deviation as abnormal; Otherwise, the image is not processed; Record the number of images marked as abnormal in each deviation-time set, determine whether the number of abnormalities is less than a set threshold N, if so, consider that the road detection unit is normal, and continue to judge the next road detection unit; otherwise, in the road detection unit where the number of abnormal images exceeds the threshold N, select all abnormal images, calculate the difference value of the attitude angle deviation of each pair of adjacent abnormal images and the time difference Δt of the corresponding moment, and then calculate the mean value of the deviation difference and the time difference E(Δt); continue judging the mean values of the deviation difference and the time difference E(Δt) whether respectively less than the respective threshold value and T Δt If yes, it is considered that the road detection unit is normal; otherwise, when any mean value is greater than or equal to the corresponding threshold value, it is considered that the road detection unit is abnormal, an alarm is given and the road number corresponding to the abnormal road detection unit and all frame images contained therein are output.

2. The low-complexity unmanned aerial-based road anomaly identification method of claim 1, wherein, The road includes: facilities with parallel line structure, such as highway, river embankment, tunnel, pipeline and railway. 3.The low-complexity unmanned aerial road anomaly identification method of claim 1, wherein, The edge feature extraction is specifically: The target road area and the background area are identified for each frame of image through semantic segmentation, the mask image of the target road area is obtained after excluding the background area, and the mask image is smoothed through a filter; Then the picture gradient is calculated through an operator, the gradient direction of each pixel is calculated, thresholding is performed, and a target road area binary image of encoded gradient direction is formed.

4. The low complexity unmanned aerial road anomaly identification method of claim 1, wherein, The vanishing point refers to: The edge straight line of the target road area binary image is extracted to obtain a plurality of straight lines, a plurality of intersection points exist among all non-collinear straight lines, the horizontal and vertical coordinates of the intersection points in the image coordinate system are respectively composed into a horizontal coordinate sequence and a vertical coordinate sequence, the median values of the horizontal and vertical coordinate sequences are respectively taken, and a new coordinate, i.e. the vanishing point coordinate, is obtained.

5. The low complexity unmanned aerial road anomaly identification method of claim 1, wherein, The road target region corresponds to a deviation Δθ of the unmanned aerial vehicle attitude angle at the vanishing point jitter of the n-th frame image n The calculation formula is: wherein, is the actual attitude angle of the UAV sensor at the time corresponding to the nth frame of image.

6. The low-complexity unmanned aerial road anomaly identification method of claim 1, wherein, the threshold value For determining whether the measured value of the attitude angle of the UAV and the estimated value have a significant deviation, when the deviation is greater than a threshold value, it is considered that the reason for the disappearance point jitter is caused by the UAV jitter.

7. The low-complexity unmanned aerial-based road anomaly identification method of claim 1, wherein, Each of the bias-time sets respectively corresponds to a road detection unit, and the ith road detection unit corresponds to a bias-time set U composed of K attitude angle biases i is:

8. The low-complexity unmanned aerial-based road anomaly identification method of claim 1, wherein, The The bias representing the pose of the UAV reaches a critical stability The value of E(Δt) when the bias representing the pose of the UAV reaches a critical stability Δt The value of E(Δt) when the bias representing the pose of the UAV reaches a critical stability

Images