CN109334563B - Anti-collision early warning method based on pedestrians and riders in front of road - Google Patents
Anti-collision early warning method based on pedestrians and riders in front of road Download PDFInfo
- Publication number
- CN109334563B CN109334563B CN201811013119.7A CN201811013119A CN109334563B CN 109334563 B CN109334563 B CN 109334563B CN 201811013119 A CN201811013119 A CN 201811013119A CN 109334563 B CN109334563 B CN 109334563B
- Authority
- CN
- China
- Prior art keywords
- early warning
- target
- image
- riders
- pedestrians
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60Q—ARRANGEMENT OF SIGNALLING OR LIGHTING DEVICES, THE MOUNTING OR SUPPORTING THEREOF OR CIRCUITS THEREFOR, FOR VEHICLES IN GENERAL
- B60Q9/00—Arrangement or adaptation of signal devices not provided for in one of main groups B60Q1/00 - B60Q7/00, e.g. haptic signalling
- B60Q9/008—Arrangement or adaptation of signal devices not provided for in one of main groups B60Q1/00 - B60Q7/00, e.g. haptic signalling for anti-collision purposes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
Abstract
The invention discloses an anti-collision early warning method based on pedestrians and riders in front of a road, belongs to the field of driving assistance systems, and comprises three aspects of environment perception, information interpretation and target state judgment. The method comprises the steps of sending a collected training set into a YOLO-R network for training, detecting and classifying targets, and realizing multi-target tracking by adopting Kalman filtering; carrying out inverse perspective transformation on the image to obtain an IPM image, fitting a relation curve of the original image and pixel coordinates of the IPM image through data regression modeling, and estimating the distance according to the linear relation of the pixel coordinates of the IPM image and world coordinates; according to the speed and the braking distance of the vehicle, an early warning activation area is determined, then whether the target is dangerous or not is judged by adopting a fuzzy early warning algorithm on the target in the activation area, and if the target is dangerous, a driver is reminded in time, so that the occurrence of accidents is effectively reduced, and the safety of pedestrians and riders is protected.
Description
Technical Field
The invention belongs to the field of active safety of automobiles, relates to knowledge of an image processing and anti-collision early warning system, and particularly relates to an anti-collision early warning method based on pedestrians and riders in front of a road.
Background
Under the typical mixed traffic environment in China, pedestrians, riders, vehicles and the like are road traffic participants, wherein the pedestrians and the riders are vulnerable groups, are exposed outside, have no protective measures, and are more difficult to guarantee the personal safety in traffic accidents, so the safety of the pedestrians and the riders is particularly important to protect.
The traditional pedestrian and rider detection algorithm is used for training a classifier by manually extracting features such as HOG, Haar, LUV and the like, and completing target detection. The method can obtain a good detection effect under specific conditions, but the detection effect of the manually designed features is not good under the conditions of dim light, obvious target posture change and complex scenes. In contrast, deep learning can extract features in an image through a convolutional layer, and the detection effect of the deep learning is obviously superior to that of a traditional machine learning method. With the enhancement of hardware computing power and the establishment of a large number of training data sets, deep learning is developed vigorously. In the aspect of target detection, from RCNN, Fast-RCNN and Fast RCNN to YOLO, SSD and YOLOv2, the speed and accuracy of target detection are greatly broken through.
Chinese patent (CN1O2765365A) discloses a pedestrian detection method and a pedestrian anti-collision early warning system based on machine vision, adopts a pedestrian classifier to detect pedestrians on roads, blurs the individual characteristics among the pedestrians, reduces the influence of individual difference on the detection result, judges the possibility of accidents through the anti-collision early warning system, but has low detection precision and neglects the safety of riders. Chinese patent (CN204870868U) discloses an automobile anti-collision and pedestrian protection early warning system based on multiple sensors, which utilizes laser ranging and ultrasonic ranging to realize early warning of front and rear vehicles, obstacles or pedestrians of an automobile.
Disclosure of Invention
Aiming at the problems, the invention provides an anti-collision early warning method based on pedestrians and riders in front of a road, which is characterized in that an anti-collision early warning system based on multi-information fusion is constructed according to the position, the transverse and longitudinal distance, the speed and the collision time TTC of a target of comprehensive analysis to judge the danger degree of the target, the pedestrians and the riders exposed outside can be effectively detected, only the target enters an activation area according to the setting of an early warning activation area, then the target state judgment is carried out, and the safety of the pedestrians and the riders is protected.
The invention adopts the following specific technical scheme:
an anti-collision early warning method based on pedestrians and riders in front of a road comprises the following steps:
s1, constructing an anti-collision early warning system based on pedestrians and riders in front of the road;
s2, performing model training by using a YOLOv2 improved network YOLO-R in offline training;
s3, inputting a frame of image into a trained YOLO-R network, detecting and classifying travelers and riders, and realizing multi-target tracking;
s4, calculating the transverse and longitudinal distance between the vehicle and the front target;
and S5, setting an early warning activation area according to the speed of the vehicle, judging whether the target is in the activation area by using the transverse and longitudinal distances, calculating an early warning index for the target in the early warning activation area, substituting the index into an early warning system, and determining the early warning level.
Further, the overall structure of the anti-collision early warning system based on pedestrians and riders in front of the road comprises three modules of environment perception, information interpretation and target state judgment, wherein the environment perception comprises target detection and tracking, vanishing point detection and vehicle speed acquisition; the target detection and tracking and the vanishing point detection are realized by acquiring related information by a front-view camera and a video acquisition card, transmitting image information to a PC (personal computer) through the video acquisition card, acquiring the vehicle speed by a GPS (global positioning system) module, reading the information and realizing a target state judgment module in the PC, and finally displaying an early warning result by a software interface.
Further, the building process of the YOLO-R network is as follows: on the basis of a YOLOv2 network structure, anchor boxes are clustered, a passhrough layer is removed, and a residual error network is added to form a YOLO-R network.
Further, the model training process in S2 is divided into two parts, namely, forward propagation and backward propagation, specifically: carrying out forward propagation calculation on the training sample, and finally outputting the relative position of the candidate frame, the confidence degree of the included target and the category probability information; continuously updating weights of each layer of the network by using a back propagation algorithm and a small batch gradient descent method, and reducing the value of a cost function; and repeating the process continuously, and finishing one iteration when all samples are trained.
Further, the algorithm in S4 includes an offline process and an online process, where the offline process specifically includes: establishing a regression model of pixel coordinates of the original image and the IPM image; the online process specifically comprises the following steps: obtaining the pixel coordinates of the middle point of the bottom edge of the target rectangular frame; calculating a pitch angle of the camera by a road vanishing point detection algorithm, and correcting the coordinates of the target pixels according to a pitch angle change value delta theta; then, obtaining an IPM image pixel coordinate corresponding to the original image pixel coordinate through a regression model; and finally, estimating the transverse and longitudinal distances of the target according to the linear relation between the pixel coordinates of the IPM image and the world coordinates.
Further, the longitudinal distance X ═ h (h)IPM-v′)·σ2+XminWherein h isIPMRepresenting the height of the image, v' representing the pixel coordinates of the IPM image, σ2Representing the actual physical distance value, X, represented by a unit pixel in the vertical directionminThe actual minimum distance in front of the IPM image.
Further, the transverse distance Y ═ u' -wIPM/2)·σ1Wherein w isIPMRepresenting the width of the image, u' representing the pixel coordinates of the IPM image, σ1Indicating the actual physical distance value represented by the unit pixel in the horizontal direction.
Further, the formula for judging whether the target is in the activation region by using the transverse and longitudinal distances is as follows:
dfminthe closest distance between the vehicle and the target, dfmaxThe furthest longitudinal distance of the vehicle from the target.
Further, the fuzzy early warning algorithm is adopted for determining the early warning grade: and determining the early warning grade, the early warning index set and the early warning weight set, and then determining the membership degree of the weight set of each index in the early warning indexes to the early warning grade, thereby determining a fuzzy evaluation matrix and determining the early warning grade as the current state of the target.
The invention has the beneficial effects that:
1. the invention uses the improved network YOLOv2 YOLO-R to detect the pedestrian and the rider, can automatically extract the higher-level features of the representation target, and improves the detection performance of the pedestrian and the rider through feature fusion between levels.
2. The monocular distance measurement algorithm based on the dynamic inverse perspective transformation of road vanishing point estimation and the data regression modeling reduces the influence of the change of the pitching angle of the camera on the distance measurement precision, and improves the distance measurement precision.
3. The invention uses a fuzzy comprehensive evaluation method to carry out anti-collision early warning, obtains a fuzzy evaluation matrix by utilizing the membership degree of a plurality of early warning indexes to the early warning grade, then comprehensively evaluates to determine the final result, determines the danger grade, reminds the driver to reduce the occurrence of accidents and ensures the safety of pedestrians and riders.
Drawings
Fig. 1 is a general structural diagram of an anti-collision early warning method based on pedestrians and riders in front of a road according to the present invention;
FIG. 2 is a block diagram of a YOLO-R network based on the improvement of the YOLOv2 network;
FIG. 3 is a diagram of a target detection and tracking algorithm of the present invention;
FIG. 4 is a flowchart of a monocular distance measuring algorithm based on inverse perspective transformation and data regression modeling according to the present invention;
FIG. 5 is a regression model of pixel ordinates of original image and IPM image in accordance with the present invention;
FIG. 6 is a regression model of the original image pixel ordinates v and Δ u of the present invention;
fig. 7 is a schematic view of the active area of the present invention.
Detailed Description
The present invention will be described in detail with reference to the accompanying drawings and embodiments.
The general structure diagram of the invention is shown in figure 1, and comprises three aspects of environment perception, information interpretation and target state judgment, wherein the external environment information of a vehicle, including road vanishing point position, front target position and tracking target position, is obtained through a camera, the vehicle speed is obtained through a GPS module, the transverse and longitudinal distance of the target is calculated through the target detection tracking result, vanishing point position information and camera calibration result, an early warning activation area is set according to the vehicle speed, whether the target is in the early warning activation area is judged through the transverse and longitudinal distance, and finally early warning indexes are calculated aiming at the target in the early warning activation area and are brought into a fuzzy early warning algorithm to determine the early warning grade.
S1, an anti-collision early warning system based on pedestrians and riders in front of a road is built, the overall structure comprises three modules of environment perception, information interpretation and target state judgment, the environment perception comprises target detection and tracking, vanishing point detection and vehicle speed acquisition, the target detection and tracking and the vanishing point detection are achieved by a front-view camera, image information is transmitted to a PC through a video acquisition card to complete image acquisition, the camera is powered by a 12V power supply, the vehicle speed is acquired by a GPS module, the information interpretation and the target state judgment module are achieved in the PC, and finally an early warning result is displayed through a software interface, and the software architecture is achieved by combining a CUDA8.0 architecture, a deep learning acceleration library cuDNN and an OpenCV2.4.10 library under a visual studio2015 development platform.
S2, off-line training is carried out by using a YOLOv2 improved network YOLO-R, and the training process of the whole model is divided into a forward propagation part and a backward propagation part;
performing blocking clustering on a target rectangular frame marked in a sample set by using a k-means clustering method, and determining the initialization specification and the number of anchor boxes; on the basis of a YOLOv2 Network structure, a passhigh layer is removed, and a Residual Network (ResNet) is added to form a YOLO-R Network; as shown in fig. 2.
The model training process is as follows:
(1) training samples are shuffled and stored in a container, and various data expansion methods are used, including rotating the image, adjusting hue, saturation, and the like. The samples are divided into a plurality of small batches, and each batch of samples are sent to the network for training, so that the number of the batch of samples can be reduced when the GPU has insufficient memory.
(2) The samples and the label information are sent into a network, forward propagation calculation is carried out, and finally the relative position of the candidate box, the confidence degree of the contained target and the class probability information are output. The picture input into the network is normalized to n multiplied by n pixels and then divided into a multiplied by a unit cells, each unit cell is provided with b anchor boxes, each target in the sample is distributed into the corresponding unit cell according to the central point position, and according to the IOU of the anchor boxes and the ground route, the anchor box with the largest IOU is selected to be responsible for the prediction of the target. After the sample passes through the YOLO-R network, outputting the predicted value of each candidate frame: (t)x,ty,tw,th,t0,p)。
(3) And continuously updating the weights of all layers of the network by using a back propagation algorithm and a small batch gradient descent method, and reducing the value of the cost function.
(4) And repeating the process continuously, and finishing one iteration when all samples are trained. The training samples are resized every 10 iterations so that the trained network can better predict pictures of different sizes. When the iteration number reaches the maximum value or the training error is not reduced for a long time, the training is stopped.
S3, as shown in fig. 3, target detection and tracking: inputting a frame of image into a trained YOLO-R network, detecting and classifying travelers and riders, and realizing multi-target tracking by adopting Kalman filtering; the specific process is as follows:
(1) inputting the image and the label information into a trained network model, dividing the image into a x a unit cells, predicting b candidate frames by each unit cell, and predicting a x b candidate frames in total; then, predicting the relative position of each candidate frame by a network forward algorithm: t is tx、ty、tw、thDegree of confidence t0And the posterior probability p of the class to which it belongs.
(2) For predicted tx、ty、tw、thAnd t0And performing mapping transformation to obtain a window which is closer to the anchor box as a detection frame.
(3) By setting a threshold T (in this embodiment, T is equal to c) of the confidence, a detection frame with a low possibility is removed, specifically: multiplying the sigma (t0) by max (p) to obtain the confidence coefficient that the detection frame belongs to a certain class; if the result is greater than the threshold T, the detection box is retained, otherwise, the detection box is removed.
(4) Respectively carrying out non-maximum suppression processing on each category and removing a redundant window, and specifically comprising the following steps: arranging the detection frames of each category according to the degree of confidence; finding out the detection frame with the highest confidence coefficient, then sequentially calculating an IOU (interaction Over Union) with other frames, deleting the frame when the IOU is larger than a threshold value d, and otherwise, keeping the frame; selecting the detection frame with the highest confidence from the unprocessed detection frames, and repeating the steps until all windows are processed; and outputting the position, the category and the confidence of the remaining detection frame.
(5) Further fusing the results output by the detection algorithm by using a matching algorithm to finish the classification of pedestrians and riders; and tracking the multiple targets by utilizing Kalman filtering.
S4, calculating the transverse and longitudinal distances through a monocular distance measurement algorithm of inverse perspective transformation and data regression modeling, wherein the whole algorithm comprises an off-line process and an on-line process, and is shown in FIG. 4;
an off-line process: firstly, a vehicle-mounted image is obtained, then an IPM (image perceptual mapping) image is obtained through camera calibration and inverse Perspective transformation, and a regression model of an original image and pixel coordinates of the IPM image is established, as shown in fig. 5.
An online process: reading in a video image in real time, and obtaining a pixel coordinate of a middle point of the bottom edge of a target rectangular frame through a detection and tracking algorithm; calculating a pitch angle of the camera by a road vanishing point detection algorithm, and correcting the coordinates of the target pixels according to a pitch angle change value delta theta; then, obtaining an IPM image pixel coordinate corresponding to the original image pixel coordinate through a regression model; and further estimating the transverse and longitudinal distances of the target according to the linear relation between the pixel coordinates of the IPM image and the world coordinates.
From a mapping curve drawn by the observation data, a relationship between v and v' is fitted, and the longitudinal distance X is obtained by equation (1).
X＝(hIPM-v′)·σ2+Xmin (1)
Wherein h isIPMRepresenting the height of the image, v' representing the pixel coordinates of the IPM image, σ2Representing the actual physical distance value, X, represented by a unit pixel in the vertical directionminThe actual front minimum distance corresponding to the IPM image;
as can be seen from the regression model of the vertical coordinates v and Δ u of the original image pixel in fig. 6, there is a clear linear relationship between the vertical coordinates v and Δ u of the original image pixel, where Δ u is a pixel value on the original image represented by a unit pixel in the horizontal direction in the IPM image; the ROI area in the inverse perspective transformation is symmetrical left and right by taking a straight line where a main point of the camera is positioned as a symmetry axis, namely the pixel value of the abscissa at the central line of the IPM image corresponds to mu of the original image0(camera intrinsic parameters). According to the condition and the fitted linear equation, the relation between the pixel coordinates (mu, nu) of the original image and the pixel abscissa u' of the IPM image can be solved, and then the transverse distance Y of the target is solved by using the formula (2).
Y＝(u′-wIPM/2)·σ1 (2)
Wherein, wIPMRepresenting the width of the image, u' representing the pixel coordinates of the IPM image, σ1Representing an actual physical distance value represented by a unit pixel in a horizontal direction;
s5, setting an early warning activation area according to the speed of the vehicle, judging whether a target is in the activation area by using the transverse and longitudinal distances, calculating early warning indexes of the target in the early warning activation area, bringing the indexes into an early warning system, and determining an early warning level;
the early warning activation region is set, and the shape of the early warning activation region is set to be trapezoidal, as shown in fig. 7. In the activation region of fig. 7, the vehicle-to-target closest distance d is calculatedfminAnd the farthest longitudinal distance dfmax。
The lateral boundaries of the activation zone are related to both vehicle speed and speed of the target; assuming that the self-vehicle does uniform motion, the running d of the self-vehicle is calculatedfminAnd dfmaxThe required time is t (d)fmin) And t (d)fmax)。
The speed u of the vehicle can be acquired by a GPS module arranged on the vehicle, and the boundary of the activation region is solved; after the activation region is determined, it is determined whether the target is within the activation region by the following formula.
Wherein X and Y represent the transverse and longitudinal distances between the target and the vehicle;
if the target is in the activation region, the early warning system uses a fuzzy algorithm to carry out early warning, and the specific process is as follows:
(1) determining an early warning grade, an early warning index set and an early warning weight set; the early warning grades are divided into three grades which are respectively safety, attention and danger, the early warning index set is E ═ { target position (E1), transverse distance (E2), vehicle speed (E3) and longitudinal distance/TTC (E4) }, and a weight set is determined by adopting a Fuzzy Analytic Hierarchy Process (FAHP); the specific operation of determining the weight set by the FAHP method comprises the following steps: constructing a fuzzy complementary judgment matrix S and solving a weight vector W;
the fuzzy complementary judgment matrix S can be obtained by:
wherein, S (i) and S (j) respectively represent the relative importance of the indexes i and j, i, j belongs to (1, …, m), and m represents the number of the early warning indexes.
Secondly, the process of obtaining the weight vector W is as follows:
the sum of each row of the matrix S is first found:
wherein r isiIs the sum of the ith row of the matrix S.
The vector R1 is obtained (R)1，…，rk，…，rm)
And then, carrying out row transformation on the vector R1 to obtain a fuzzy consistent matrix R2:
the expression for each element in the fuzzy consensus matrix R2 is:
the other elements in each row except the main diagonal element in R2 are summed:
the sum of the elements of the fuzzy consensus matrix R2 other than the main diagonal is:
wherein liIndicates the degree of importance of index i to index i-1, toiNormalization operation, which can calculate the weight of each early warning grade; weight wiCan be expressed as formula (9), and the final weight result W is expressed as formula (10).
W＝(1/8,5/24,3/8,7/24) (10)
(2) And obtaining a comment set of each index according to a fuzzy method, wherein the position comment set of the target is { close, keep and far }, the comment set of the transverse distance is { short, medium and long }, the comment set of the speed of the vehicle is { low speed, medium speed and high speed }, and the comment set of the longitudinal distance/TTC is { short/small, medium and long/large }.
(3) After the early warning index set and the comment set thereof are established, the membership degree of the comment set of each index in the early warning index to the early warning grade needs to be determined, so that the fuzzy evaluation matrix is determined. The membership function adopts a discrete quantization form, obtains a membership table shown in table 1 according to an expert experience method, and obtains a fuzzy evaluation matrix R through table lookup.
TABLE 1 membership Table
(4) Selecting M (·, +) operator to synthesize the weight vector and the fuzzy evaluation matrix to obtain the membership vector of the target state to the early warning level
Selecting S according to the principle of maximum membership degree1And taking the early warning grade corresponding to the medium maximum value as the current state of the target.
The above-mentioned embodiments are provided to illustrate the present invention, but the present invention is not limited thereto, and any person skilled in the art should be able to make modifications and changes without departing from the spirit and scope of the present invention, and the scope of the present invention is defined by the scope of the appended claims.
Claims (8)
1. An anti-collision early warning method based on pedestrians and riders in front of a road is characterized by comprising the following steps:
s1, constructing an anti-collision early warning system based on pedestrians and riders in front of the road;
s2, performing model training by using a YOLOv2 improved network YOLO-R in offline training;
the network YOLO-R is specifically as follows: on the basis of a YOLOv2 network structure, clustering and selecting anchor boxes, removing a passhrough layer, and adding a residual network to form a YOLO-R network;
s3, inputting a frame of image into a trained YOLO-R network, detecting and classifying travelers and riders, and realizing multi-target tracking;
s4, calculating the transverse and longitudinal distance between the vehicle and the front target;
and S5, setting an early warning activation area according to the speed of the vehicle, judging whether the target is in the activation area by using the transverse and longitudinal distances, calculating an early warning index for the target in the early warning activation area, substituting the index into an early warning system, and determining the early warning level.
2. The anti-collision early warning method based on the pedestrians and the riders in the front of the road according to claim 1, wherein the overall structure of the anti-collision early warning system based on the pedestrians and the riders in the front of the road comprises three modules of environment perception, information interpretation and target state judgment, wherein the environment perception comprises target detection and tracking, vanishing point detection and vehicle speed acquisition; the target detection and tracking and the vanishing point detection are realized by acquiring related information by a front-view camera, transmitting image information to a computer by a video acquisition card, acquiring vehicle speed information by a GPS module, reading the information and realizing a target state judgment module in the computer, and finally displaying an early warning result by a software interface.
3. The anti-collision early warning method based on pedestrians and riders in front of the road according to claim 1, wherein the model training process in S2 is divided into two parts, namely forward propagation and backward propagation, specifically: carrying out forward propagation calculation on the training sample, and finally outputting the relative position of the candidate frame, the confidence degree of the included target and the category probability information; continuously updating weights of each layer of the network by using a back propagation algorithm and a small batch gradient descent method, and reducing the value of a cost function; and repeating the process continuously, and finishing one iteration when all samples are trained.
4. The anti-collision early warning method based on pedestrians and riders in front of the road according to claim 1, wherein the algorithm in S4 includes two parts, namely an off-line process and an on-line process, and the off-line process specifically includes: establishing a regression model of pixel coordinates of the original image and the IPM image; the online process specifically comprises the following steps: obtaining the pixel coordinates of the middle point of the bottom edge of the target rectangular frame; calculating a pitch angle of the camera by a road vanishing point detection algorithm, and correcting the coordinates of the target pixels according to a pitch angle change value delta theta; then, obtaining an IPM image pixel coordinate corresponding to the original image pixel coordinate through a regression model; and finally, estimating the transverse and longitudinal distances of the target according to the linear relation between the pixel coordinates of the IPM image and the world coordinates.
5. The method as claimed in claim 4, wherein the longitudinal distance X is (h) in the method for warning collision avoidance based on pedestrians and riders ahead of roadIPM-v′)·σ2+XminWherein h isIPMRepresenting the height of the image, v' representing the pixel coordinates of the IPM image, σ2Representing the actual physical distance value, X, represented by a unit pixel in the vertical directionminThe actual minimum distance in front of the IPM image.
6. The method as claimed in claim 4, wherein the transverse distance Y is (u' -w) ═ wIPM/2)·σ1Wherein w isIPMRepresenting the width of the image, u' representing the pixel coordinates of the IPM image, σ1Indicating the actual physical distance value represented by the unit pixel in the horizontal direction.
7. The method for warning collision avoidance based on pedestrians and riders in front of road according to claim 5 or 6, wherein the formula for determining whether the target is in the activation area by using the transverse and longitudinal distance is as follows:
dfminthe closest distance between the vehicle and the target, dfmaxThe furthest longitudinal distance of the vehicle from the target.
8. The anti-collision early warning method based on the pedestrians and the riders in front of the road according to claim 1, wherein a fuzzy early warning algorithm is adopted for determining the early warning level: and determining the early warning grade, the early warning index set and the early warning weight set, and then determining the membership degree of the comment set of each index in the early warning index to the early warning grade, thereby determining a fuzzy evaluation matrix and determining the early warning grade as the current state of the target.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811013119.7A CN109334563B (en) | 2018-08-31 | 2018-08-31 | Anti-collision early warning method based on pedestrians and riders in front of road |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811013119.7A CN109334563B (en) | 2018-08-31 | 2018-08-31 | Anti-collision early warning method based on pedestrians and riders in front of road |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109334563A CN109334563A (en) | 2019-02-15 |
| CN109334563B true CN109334563B (en) | 2021-06-22 |
Family
ID=65296814
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811013119.7A Active CN109334563B (en) | 2018-08-31 | 2018-08-31 | Anti-collision early warning method based on pedestrians and riders in front of road |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109334563B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109919074B (en) * | 2019-03-01 | 2022-11-04 | 中国科学院合肥物质科学研究院 | Vehicle sensing method and device based on visual sensing technology |
| CN110929606A (en) * | 2019-11-11 | 2020-03-27 | 浙江鸿泉车联网有限公司 | Vehicle blind area pedestrian monitoring method and device |
| CN113031580A (en) * | 2019-12-25 | 2021-06-25 | 奥迪股份公司 | Remote assistance system, control device, corresponding server, vehicle and method |
| CN111860097B (en) * | 2020-04-16 | 2024-02-20 | 西安电子科技大学 | Abnormal behavior detection method based on fuzzy theory |
| CN111783700B (en) * | 2020-07-06 | 2023-11-24 | 中国交通通信信息中心 | Automatic recognition and early warning method and system for pavement foreign matters |
| CN111921199B (en) * | 2020-08-25 | 2023-09-26 | 腾讯科技（深圳）有限公司 | Method, device, terminal and storage medium for detecting state of virtual object |
| CN112249021B (en) * | 2020-10-14 | 2022-06-24 | 广州汽车集团股份有限公司 | Road pedestrian collision risk prediction method and system |
| CN112349144B (en) * | 2020-11-10 | 2022-04-19 | 中科海微(北京)科技有限公司 | Monocular vision-based vehicle collision early warning method and system |
| CN113192331B (en) * | 2021-04-26 | 2022-04-15 | 吉林大学 | Intelligent early warning system and early warning method for riding safety in internet environment |
| CN113449589B (en) * | 2021-05-16 | 2022-11-15 | 桂林电子科技大学 | Method for calculating driving strategy of unmanned vehicle in urban traffic scene |
| CN114267181B (en) * | 2022-03-03 | 2022-06-03 | 深圳市旗扬特种装备技术工程有限公司 | Pedestrian early warning method and device for urban road intersection and electronic equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102765365A (en) * | 2011-05-06 | 2012-11-07 | 香港生产力促进局 | Pedestrian detection method based on machine vision and pedestrian anti-collision warning system based on machine vision |
| CN106156725A (en) * | 2016-06-16 | 2016-11-23 | 江苏大学 | A kind of method of work of the identification early warning system of pedestrian based on vehicle front and cyclist |
| US9760806B1 (en) * | 2016-05-11 | 2017-09-12 | TCL Research America Inc. | Method and system for vision-centric deep-learning-based road situation analysis |
| CN107862287A (en) * | 2017-11-08 | 2018-03-30 | 吉林大学 | A kind of front zonule object identification and vehicle early warning method |
| CN107972662A (en) * | 2017-10-16 | 2018-05-01 | 华南理工大学 | To anti-collision warning method before a kind of vehicle based on deep learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7613569B2 (en) * | 2005-07-19 | 2009-11-03 | Toyota Motor Engineering & Manufacturing North America, Inc. | Crash prediction network with visual input for vehicle |
-
2018
- 2018-08-31 CN CN201811013119.7A patent/CN109334563B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102765365A (en) * | 2011-05-06 | 2012-11-07 | 香港生产力促进局 | Pedestrian detection method based on machine vision and pedestrian anti-collision warning system based on machine vision |
| US9760806B1 (en) * | 2016-05-11 | 2017-09-12 | TCL Research America Inc. | Method and system for vision-centric deep-learning-based road situation analysis |
| CN107368890A (en) * | 2016-05-11 | 2017-11-21 | Tcl集团股份有限公司 | A kind of road condition analyzing method and system based on deep learning centered on vision |
| CN106156725A (en) * | 2016-06-16 | 2016-11-23 | 江苏大学 | A kind of method of work of the identification early warning system of pedestrian based on vehicle front and cyclist |
| CN107972662A (en) * | 2017-10-16 | 2018-05-01 | 华南理工大学 | To anti-collision warning method before a kind of vehicle based on deep learning |
| CN107862287A (en) * | 2017-11-08 | 2018-03-30 | 吉林大学 | A kind of front zonule object identification and vehicle early warning method |
Non-Patent Citations (4)
| Title |
|---|
| Pedestrian Detection Based on YOLO Network Model;Wenbo Lan等;《Proceedings of 2018 IEEE,International Conference on Mechatronics and Automation》;20180808;第1547-1551页 * |
| 基于OpenCV的前方车辆检测和前撞预警;刘军等;《汽车技术》;20170630(第6期);第11-16页 * |
| 基于YOLO网络的行人检测方法;高宗等;《计算机工程》;20180531;第44卷(第5期);第215-219页 * |
| 基于单目视觉车辆姿态角估计和逆透视变换的车距测量;刘军等;《农业工程学报》;20180731;第34卷(第13期);第70-76页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109334563A (en) | 2019-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109334563B (en) | Anti-collision early warning method based on pedestrians and riders in front of road | |
| CN110007675B (en) | Vehicle automatic driving decision-making system based on driving situation map and training set preparation method based on unmanned aerial vehicle | |
| CN113366486B (en) | Object classification using out-of-region context | |
| US9767368B2 (en) | Method and system for adaptive ray based scene analysis of semantic traffic spaces and vehicle equipped with such system | |
| CN108638999B (en) | Anti-collision early warning system and method based on 360-degree look-around input | |
| JP6574611B2 (en) | Sensor system for obtaining distance information based on stereoscopic images | |
| US20170300763A1 (en) | Road feature detection using a vehicle camera system | |
| EP3623990B1 (en) | Methods and systems for computer-based determination of presence of objects | |
| Guo et al. | A multimodal ADAS system for unmarked urban scenarios based on road context understanding | |
| CN111369617B (en) | 3D target detection method of monocular view based on convolutional neural network | |
| US20230213643A1 (en) | Camera-radar sensor fusion using local attention mechanism | |
| CN114495064A (en) | Monocular depth estimation-based vehicle surrounding obstacle early warning method | |
| EP3842307A1 (en) | System and method for providing vehicle safety distance and speed alerts under slippery road conditions | |
| CN111091038A (en) | Training method, computer readable medium, and method and apparatus for detecting vanishing points | |
| CN111950483A (en) | Vision-based vehicle front collision prediction method | |
| CN110991264A (en) | Front vehicle detection method and device | |
| Alpar et al. | Intelligent collision warning using license plate segmentation | |
| JP5434861B2 (en) | Moving object detection apparatus and program | |
| CN116012822B (en) | Fatigue driving identification method and device and electronic equipment | |
| CN117115690A (en) | Unmanned aerial vehicle traffic target detection method and system based on deep learning and shallow feature enhancement | |
| CN116311136A (en) | Lane line parameter calculation method for driving assistance | |
| Aboah et al. | Ai-based framework for understanding car following behaviors of drivers in a naturalistic driving environment | |
| WO2023009180A1 (en) | Lidar-based object tracking | |
| US20220129685A1 (en) | System and Method for Determining Object Characteristics in Real-time | |
| CN116778262B (en) | Three-dimensional target detection method and system based on virtual point cloud |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |