CN110097571B - Quick high-precision vehicle collision prediction method - Google Patents
Quick high-precision vehicle collision prediction method Download PDFInfo
- Publication number
- CN110097571B CN110097571B CN201910350810.2A CN201910350810A CN110097571B CN 110097571 B CN110097571 B CN 110097571B CN 201910350810 A CN201910350810 A CN 201910350810A CN 110097571 B CN110097571 B CN 110097571B
- Authority
- CN
- China
- Prior art keywords
- motion
- interaction field
- vehicle collision
- collision prediction
- motion interaction
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/207—Analysis of motion for motion estimation over a hierarchy of resolutions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
- G06T7/251—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
A fast and high-precision vehicle collision prediction method comprises the following steps: firstly, generating a motion interaction field for a video frame input at a certain moment by using a Farnesback algorithm and a Gaussian difference kernel function, expressing the symmetry of the motion interaction field by using the area difference of positive and negative areas, determining an abnormal value at the moment by adopting a linear regression model, and finally predicting whether a vehicle collides in a hard threshold mode.
Description
Technical Field
The invention belongs to the field of intelligent traffic, and relates to a rapid and high-precision vehicle collision prediction method which is particularly suitable for predicting road vehicle collision in real time and all weather.
Background
Many research groups at home and abroad try to study an automatic detection method of a vehicle collision accident through computer vision and pattern recognition technology. SiyuXia et al propose a method based on low-rank matrix approximation to rapidly and effectively automatically measure traffic accidents in video. It first divides each frame into non-overlapping blocks associated with different weights, then extracts the motion matrix of the video segment, and after using a low rank matrix approximation to associate a normal traffic scene with a set of motion subspaces, identifies traffic incidents at the moment of increasing approximation errors. Yu Chen et al propose a classifier-based monitoring method by observing the last second of video before a collision of a motor vehicle as a detection target. It first extracts local motion information features from the time domain using the OF-SIFT algorithm for motion detection, and then classifies traffic accidents using an Extreme Learning Machine (ELM) classifier as a basic classifier. Li Yuee and the like are used for obtaining a traffic accident probability formula by extracting parameters such as smoke, fragments and change of the speed of a moving target in a traffic monitoring video and carrying out formula fitting by combining with a maquardt method, calculating the accident probability, and judging whether the vehicle collides according to the probability. The baya et al analyzes some specific states of the vehicle in the video, such as size, position, etc., and judges whether the two targets have overlapping areas according to the movements, thereby judging whether the vehicle collides. Liu Weiqi and the like calculate a bounding box of a vehicle by using an OBB algorithm, then project a ground plane of the bounding box to obtain a rectangular two-dimensional bounding box of the vehicle, and then detect whether the rectangles intersect on the same plane so as to judge whether the vehicle collides. Yin Chune and the like propose an adaptive threshold highway traffic accident detection method based on a wavelet equation, and the threshold is dynamically and adaptively adjusted based on the change of traffic running states, so that a relatively high accident detection rate is achieved. Yin Hongpeng by intelligent analysis, detecting the motion of the vehicle by background difference method, tracking the motion vehicle by mean shift algorithm, and weighting and fusing the obtained parameters such as vehicle position, speed and motion direction to detect the vehicle behavior (collision, stop running, etc. of the vehicle)
The algorithm well realizes traffic accident detection, but has the problems of inaccurate target detection, poor tracking effect, low collision discrimination accuracy and the like.
Atev and the like adopt a plurality of cameras to monitor the same intersection at the same time, acquire a motion state vector of a target after data fusion, and establish a vehicle collision prediction algorithm by utilizing a vehicle track extrapolation assumption for realizing collision detection. Lauresh et al have attempted to establish a set of severity indicators that can measure traffic interaction events throughout the process, including distance collision time (TTC), time of dominance (TAdv), gap Time (TG), speed, etc., based on which traffic collisions are classified and identified according to the severity of the event throughout the process. Ismail and the like detect, track and classify pedestrians and vehicles in traffic scenes, automatically calculate severity indexes by means of extrapolation assumptions, and identify collision between the pedestrians and the vehicles accordingly. Zhang Fangfang microscopic traffic parameters such as track, speed, acceleration and the like of the collision vehicles are extracted from the video image sequence, a majority of traffic collision judging methods are established, and automatic detection of motor vehicle traffic collision videos under simple traffic conditions is primarily realized. Qu Zhaowei and the like are used for predicting the moving target track by introducing a GM (1, 1) gray model on the basis of moving target detection and tracking, and realizing automatic judgment of traffic conflict at a signalless intersection on the basis of establishing two conditions of traffic conflict judgment. Hu and the like learn the motion mode of the vehicle by using a fuzzy self-organizing neural network algorithm, match and predict the motion trail of the vehicle according to the motion mode, and finally predict the accident by calculating the occurrence probability of collision between vehicles. Saunier et al set up a video-based automatic analysis method of cross security, learn the track prototype through the longest public subsequence (LCSS) algorithm, calculate the collision possibility between vehicles on this basis, judge the traffic conflict.
Although research on traffic conflict technology based on videos has achieved some staged results, most of the existing conflict judging methods simply and directly apply microscopic traffic data collected by videos, and are often finally attributed to the traditional traffic conflict judging thought.
The patent aims to provide the rapid and high-precision vehicle collision prediction method which has the advantages of small calculated amount, high calculation speed, strong noise resistance and high detection precision and can predict the vehicle collision traffic accidents in real time all weather.
Disclosure of Invention
1. A rapid and high-precision vehicle collision prediction method is characterized by comprising the following steps:
step one, constructing a motion interaction field F 2 (x, y) according to the following steps:
firstly, inputting a video frame, and performing downsampling or upsampling on an original image frame to obtain an image frame f (x, y) with the size of MxN;
secondly, extracting the motion information of a target object in the video sequence by using a farnebback optical flow algorithm: processing the ith pixel value f (x i ,y i ) Obtaining the velocity in the x direction
And speed in y direction>
Thereby obtaining an optical flow field V with the size of MxN 0x And V 0y ；
Thirdly, downsampling an optical flow field to obtain optical flow fields Vx and Vy with the size of M/10 XN/10;
fourth, vx and Vy are applied to kernel function K to obtain motion interaction field F with the size of M/10 XN/10 1 (x, y) having the expression:
wherein K is a kernel function, and the expression is:
in the above-mentioned method, the step of,
is forward center position +.>
For the rearward center position, k is the width sigma and the center is (x c ,y c ) The expression of the Gaussian kernel function is as follows:
fifth step, for F 1 (x, y) obtaining a motion interaction field F with the size of 2 Mx 2N by bicubic interpolation 2 (x,y)。
Step two, obtaining the abnormal value of the motion interaction field at the time t by using the area difference method
The method comprises the following steps:
first, utilize threshold th 1 The motion interaction field F obtained in the last step is obtained 2 The absolute value of the pixel value in (x, y) is greater than th 1 Sum S of positive and negative area 1 、S 2 ；
Second, calculating the absolute value M of the positive and negative area difference abn The expression is:
M abn ＝|S 1 -S 2 |；
third step, smoothing M using autoregressive model abn Obtaining
The autoregressive model is as follows:
where α is an attenuation parameter indicating the propagation degree, th 2 Is a fluctuation threshold;
an outlier representing the motion interaction field at time t;
step three, utilizing the abnormal value of the motion interaction field obtained in the step two
Setting threshold th 3 The vehicle collision is predicted, and the expression is:
wherein L is t Is the final label of the t frame anomaly. If it is
Judging that collision occurs, and carrying out early warning; otherwise, it is determined that the vehicle is not involved in a collision.
2. The rapid and high-precision vehicle collision prediction method according to claim 1, characterized in that: in the second and third steps, the detection precision is highest when the attenuation parameter alpha is 0.8 through experimental analysis.
3. The rapid and high-precision vehicle collision prediction method according to claim 1, characterized in that: in the third step, the abnormal value obtained by the area difference is determined, and the lower the set early warning threshold value is, the more sensitive to the vehicle collision prediction is, but the accuracy of the collision prediction is reduced, and the abnormal value is set to be about 1/3 of the collision threshold value in the experiment.
The invention has the remarkable effects that: the method has the advantages of small calculated amount, high calculated speed, strong noise immunity and high detection precision, and can be used for carrying out real-time vehicle collision prediction and detection on the vehicle collision phenomenon in the video with higher accuracy.
Drawings
FIG. 1 is a flow chart of a vehicle collision prediction algorithm;
FIG. 2 is a video frame and its motion interaction field of a toy car simulating normal traffic;
FIG. 3 is a video frame and its motion interaction field of a toy car simulating abnormal traffic;
fig. 4 is an outlier variation of a video motion interaction field containing a collision frame.
Detailed Description
The invention will be described in further detail with reference to the drawings and the specific examples.
Fig. 1 is a flowchart of an algorithm according to the present invention.
The parameters in the algorithm are set as follows:
(A) M N is typically 240X 320;
(B)α＝0.8；
(C)th 1 ＝55；
(D)th 2 ＝300；
(E)th 3 ＝1100；
(F)σ＝10；
a video is input, the left side of fig. 2 is a frame which is taken from a car simulated traffic accident video and is not collided with, and the left side of fig. 3 is a frame which is taken from the same video and is likely to collide with the vehicle. Downsampling an image frame in the video results in an image frame f (x, y) having an mxn value of 240×320. Then the algorithm obtains motion information by using a Farnesback optical flow method: the velocity in the x-direction and the velocity in the y-direction, thereby obtaining an optical flow field V of 240X 320 0x And V 0y The method comprises the steps of carrying out a first treatment on the surface of the Downsampling the optical flow field results in optical flow fields Vx and Vy of size 24 x 32.
Using the resulting Vx and Vy, the kernel function is carried in:
by passing through
Obtaining a motion interaction field F with the size of 24 multiplied by 32 1 (x, y), then to F 1 (x, y) Using bicubic interpolation to obtain a largeAs small as 2M x 2N motion interaction field F 2 (x, y). As shown in the right side of fig. 2 and the right side of fig. 3, the motion interaction field of a frame in which the vehicle is not yet crashed and the motion interaction field of a frame in which the vehicle is likely to crash are taken from the same video.
Using the determined motion interaction fields F respectively 2 The absolute value of the pixel value in (x, y) is greater than th 1 Sum S of positive and negative area 1 、S 2 By M abn ＝|S 1 -S 2 Computing outlier M abn . Then smoothing M using an auto-regressive model with attenuation parameter α=0.8 abn Obtaining abnormal values of two frames of video
Wherein the method comprises the steps of
At this time by an abnormal value
A vehicle collision is predicted.
FIG. 4 is an outlier per frame of a selected video segment, it being seen that the outlier of the uncorrupted frames is less than the threshold th 3 =1100, whereas the collision frame is larger than it.
As can be seen from fig. 4: with the pre-warning threshold 300 set for a possible collision, wherein frame 176 is the collision frame, and frame 172 and before is the normal frame. The threshold is designed, and 4 frames of predictions are advanced to collide.
Claims (3)
1. A rapid and high-precision vehicle collision prediction method is characterized by comprising the following steps:
step one, constructing a motion interaction field F 2 (x, y) according to the following steps:
firstly, inputting a video frame, and performing downsampling or upsampling on an original image frame to obtain an image frame f (x, y) with the size of MxN;
secondly, extracting the motion information of a target object in the video sequence by using a farnebback optical flow algorithm: processing the ith pixel value f (x i ,y i ) Obtaining the velocity in the x direction
And speed in y direction>
Thereby obtaining an optical flow field V with the size of MxN 0x And V 0y ；
Thirdly, downsampling an optical flow field to obtain optical flow fields Vx and Vy with the size of M/10 XN/10;
fourth, vx and Vy are applied to kernel function K to obtain motion interaction field F with the size of M/10 XN/10 1 (x, y) having the expression:
wherein K is a kernel function, and the expression is:
in the above-mentioned method, the step of,
is forward center position +.>
For the rearward center position, k is the width sigma and the center is (x c ,y c ) The expression of the Gaussian kernel function is as follows:
fifth step, for F 1 (x, y) obtaining a motion interaction field F with the size of 2 Mx 2N by bicubic interpolation 2 (x,y)；
Step two, obtaining the abnormal value of the motion interaction field at the time t by using the area difference method
The method comprises the following steps:
first, utilize threshold th 1 The motion interaction field F obtained in the last step is obtained 2 The absolute value of the pixel value in (x, y) is greater than th 1 The sum of the positive area of (2) is S 1 The sum of the areas of the negative areas is S 2 ；
Second, calculating the absolute value M of the positive and negative area difference abn The expression is:
M abn ＝|S 1 -S 2 |
third step, smoothing M using autoregressive model abn Obtaining
The autoregressive model is as follows:
where α is an attenuation parameter indicating the propagation degree, th 2 For the fluctuation threshold value,
an outlier representing the motion interaction field at time t;
step three, utilizing the abnormal value of the motion interaction field obtained in the step two
Setting threshold th 3 The vehicle collision is predicted, and the expression is:
wherein L is t Is the final label of t frame abnormality; if it is
Judging that collision occurs, and carrying out early warning; otherwise, it is determined that the vehicle is not involved in a collision.
2. The rapid and high-precision vehicle collision prediction method according to claim 1, characterized in that: in the second and third steps, the detection precision is highest when the attenuation parameter alpha is 0.8 through experimental analysis.
3. The rapid and high-precision vehicle collision prediction method according to claim 1, characterized in that: in the third step, the abnormal value obtained by the area difference is judged, and the lower the set early warning threshold value is, the more sensitive to the vehicle collision prediction is, but the accuracy of the collision prediction is reduced, and the lower the set early warning threshold value is, in the experiment, 1/3 of the collision threshold value is set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910350810.2A CN110097571B (en) | 2019-04-28 | 2019-04-28 | Quick high-precision vehicle collision prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910350810.2A CN110097571B (en) | 2019-04-28 | 2019-04-28 | Quick high-precision vehicle collision prediction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110097571A CN110097571A (en) | 2019-08-06 |
| CN110097571B true CN110097571B (en) | 2023-06-13 |
Family
ID=67446218
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910350810.2A Active CN110097571B (en) | 2019-04-28 | 2019-04-28 | Quick high-precision vehicle collision prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110097571B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110660222B (en) * | 2019-11-01 | 2024-05-10 | 河北工业大学 | Intelligent environment-friendly electronic snapshot system for black-smoke road vehicle |
| CN111832495A (en) * | 2020-07-17 | 2020-10-27 | 中通服咨询设计研究院有限公司 | Method for detecting vehicle accident in video |
| CN115230694A (en) * | 2022-09-20 | 2022-10-25 | 毫末智行科技有限公司 | Obstacle recognition method and device for automatic driving vehicle and vehicle |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2014067276A (en) * | 2012-09-26 | 2014-04-17 | Fuji Heavy Ind Ltd | Collision determination device for vehicle |
| CN105405297A (en) * | 2015-12-21 | 2016-03-16 | 电子科技大学 | Traffic accident automatic detection method based on monitoring video |
| CN105574552A (en) * | 2014-10-09 | 2016-05-11 | 东北大学 | Vehicle ranging and collision early warning method based on monocular vision |
| CN106203267A (en) * | 2016-06-28 | 2016-12-07 | 成都之达科技有限公司 | Vehicle collision avoidance method based on machine vision |
| US9834139B1 (en) * | 2016-09-30 | 2017-12-05 | Plk Technologies Co., Ltd. | Vehicle collision warning prevention method using optical flow analysis |
| CN108389430A (en) * | 2018-01-12 | 2018-08-10 | 南京理工大学 | A kind of intersection pedestrian based on video detection and collision of motor-driven vehicles prediction technique |
| CN108846322A (en) * | 2018-05-25 | 2018-11-20 | 重庆大学 | The vehicle collision detection method of quick high accuracy |
| CN108922247A (en) * | 2018-07-25 | 2018-11-30 | 重庆大学 | A kind of ship based on AIS-navigation mark Risk-Degree of Collision evaluation method |
| CN109409238A (en) * | 2018-09-28 | 2019-03-01 | 深圳市中电数通智慧安全科技股份有限公司 | A kind of obstacle detection method, device and terminal device |
| WO2019044625A1 (en) * | 2017-09-01 | 2019-03-07 | 株式会社村上開明堂 | Collision prediction device, collision prediction method, and program |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170344855A1 (en) * | 2016-05-24 | 2017-11-30 | Agt International Gmbh | Method of predicting traffic collisions and system thereof |
-
2019
- 2019-04-28 CN CN201910350810.2A patent/CN110097571B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2014067276A (en) * | 2012-09-26 | 2014-04-17 | Fuji Heavy Ind Ltd | Collision determination device for vehicle |
| CN105574552A (en) * | 2014-10-09 | 2016-05-11 | 东北大学 | Vehicle ranging and collision early warning method based on monocular vision |
| CN105405297A (en) * | 2015-12-21 | 2016-03-16 | 电子科技大学 | Traffic accident automatic detection method based on monitoring video |
| CN106203267A (en) * | 2016-06-28 | 2016-12-07 | 成都之达科技有限公司 | Vehicle collision avoidance method based on machine vision |
| US9834139B1 (en) * | 2016-09-30 | 2017-12-05 | Plk Technologies Co., Ltd. | Vehicle collision warning prevention method using optical flow analysis |
| WO2019044625A1 (en) * | 2017-09-01 | 2019-03-07 | 株式会社村上開明堂 | Collision prediction device, collision prediction method, and program |
| CN108389430A (en) * | 2018-01-12 | 2018-08-10 | 南京理工大学 | A kind of intersection pedestrian based on video detection and collision of motor-driven vehicles prediction technique |
| CN108846322A (en) * | 2018-05-25 | 2018-11-20 | 重庆大学 | The vehicle collision detection method of quick high accuracy |
| CN108922247A (en) * | 2018-07-25 | 2018-11-30 | 重庆大学 | A kind of ship based on AIS-navigation mark Risk-Degree of Collision evaluation method |
| CN109409238A (en) * | 2018-09-28 | 2019-03-01 | 深圳市中电数通智慧安全科技股份有限公司 | A kind of obstacle detection method, device and terminal device |
Non-Patent Citations (5)
| Title |
|---|
| Implementation and performance evaluation of optical flow navigation system under specific conditions for a flying robot;Ali Aminzadeh等;《IEEE Aerospace and Electronic Systems Magazine 》;20181130;第33卷(第11期);全文 * |
| Pedestrian Collision Warning of Advanced Driver Assistance Systems;Ying-Che Kuo等;《2016 International Symposium on Computer, Consumer and Control (IS3C)》;20160818;全文 * |
| 基于HEVC的车辆异常事件检测;常同伟等;《数据采集与处理》;20180315(第02期);全文 * |
| 基于区域光流特征的异常行为检测;杜鉴豪等;《浙江大学学报(工学版)》;20110715(第07期);全文 * |
| 智能视频分析的车辆异常行为检测方法;尹宏鹏等;《重庆大学学报》;20160615(第03期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110097571A (en) | 2019-08-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108725440B (en) | Forward collision control method and apparatus, electronic device, program, and medium | |
| CN112700470B (en) | Target detection and track extraction method based on traffic video stream | |
| CN110097571B (en) | Quick high-precision vehicle collision prediction method | |
| CN105744232A (en) | Method for preventing power transmission line from being externally broken through video based on behaviour analysis technology | |
| CN110077398B (en) | Risk handling method for intelligent driving | |
| CN106652465A (en) | Method and system for identifying abnormal driving behavior on road | |
| Huang et al. | Automatic moving object extraction through a real-world variable-bandwidth network for traffic monitoring systems | |
| CN113762473B (en) | Complex scene driving risk prediction method based on multi-time space diagram | |
| CN110598511A (en) | Method, device, electronic equipment and system for detecting green light running event | |
| CN107031661A (en) | A kind of lane change method for early warning and system based on blind area camera input | |
| Rajendar et al. | Prediction of stopping distance for autonomous emergency braking using stereo camera pedestrian detection | |
| Ghahremannezhad et al. | A real time accident detection framework for traffic video analysis | |
| CN114446046A (en) | LSTM model-based weak traffic participant track prediction method | |
| CN103177248A (en) | Rapid pedestrian detection method based on vision | |
| Khan | Estimating Speeds and Directions of Pedestrians in Real-Time Videos: A solution to Road-Safety Problem. | |
| Karim et al. | An attention-guided multistream feature fusion network for localization of risky objects in driving videos | |
| CN113538968B (en) | Method and apparatus for outputting information | |
| Detzer et al. | Analysis of traffic safety for cyclists: The automatic detection of critical traffic situations for cyclists | |
| Yu et al. | Vehicle forward collision warning based upon low frequency video data: A hybrid deep learning modeling approach | |
| Tschürtz et al. | System of systems safety analysis and evaluation in ZalaZONE | |
| Aghav et al. | Deep Learning for Real Time Collision Detection and Avoidance | |
| Gao et al. | An In-Car Objects Detection Algorithm Based on Improved Spatial-Temporal Entropy Image | |
| Yang et al. | An Improved Object Detection and Trajectory Prediction Method for Traffic Conflicts Analysis | |
| Teng | Prevention Detection for Cyclists based on Faster R-CNN | |
| Hnoohom et al. | The video-based safety methodology for pedestrian crosswalk safety measured: The case of Thammasat University, Thailand |
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 |