CN103632376A - Method for suppressing partial occlusion of vehicles by aid of double-level frames - Google Patents
Method for suppressing partial occlusion of vehicles by aid of double-level frames Download PDFInfo
- Publication number
- CN103632376A CN103632376A CN201310674807.9A CN201310674807A CN103632376A CN 103632376 A CN103632376 A CN 103632376A CN 201310674807 A CN201310674807 A CN 201310674807A CN 103632376 A CN103632376 A CN 103632376A
- Authority
- CN
- China
- Prior art keywords
- occlusion
- vehicle
- blocking
- cut
- convex closure
- 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.)
- Pending
Links
Images
Abstract
The invention discloses a method for suppressing partial occlusion of vehicles by the aid of double-level frames. The method includes dividing the occlusion of foreground regions with the occlusion into weak occlusion and strong occlusion according to occlusion degrees, searching dividing lines with the optimal area ratios for extracted foreground convex hulls for the weak occlusion by an intra-frame-level process for handling occlusion on the basis of analysis on convex hulls of foreground contours, and acquiring dividing results used as occlusion separation results; suppressing the strong occlusion by a tracking-level process for handling occlusion on the basis of online sample classification. Test samples and training samples are respectively extracted from occlusion foregrounds of current frames and historical images of the vehicles without the occlusion by a sliding window process, and the historical images correspond to the occlusion foregrounds of the current frames. The test samples and the training samples are matched with one another by a KNN (K nearest neighborhood) process, the test samples are marked by labels of the training samples which are optimally matched with the test samples, and marking results correspond to the occlusion dividing results. The method has the advantages of high detection precision and adaptability and low complexity.
Description
Technical field
The invention belongs to machine vision and image processing field, the vehicle sections that relates in particular to two-stage framed structure in a kind of video detection system blocks removing method.
Background technology
In video detection system, multiple goal extraction accurately and tracking technique can obtain useful transport information such as comprising type of vehicle, track, flow and lane occupancy ratio, the detection that these information are traffic events provides a large amount of basic datas, to improving the efficiency of traffic administration and the performance of traffic safety, is bringing into play key effect.Yet, Vision imaging system is when being projected to the 2D plane of delineation by 3D scene, lost depth information, cause essence at the object of a plane, not to be projected to a plane, this makes in the target recognition and tracking process based on image, block and can often occur, especially between target, mutually block the accuracy that will have a strong impact on DETECTION OF TRAFFIC PARAMETERS.Therefore, must set up a kind of reliable, practical occlusion detection and eliminate mechanism.
At present, occlusion detection and removing method are roughly divided into four classes: the method based on N dimension auto model, the method based on characteristic model, method and the method based on inference pattern based on statistical model.Wherein, the method based on N dimension auto model utilizes N dimension (3D, 2.5D) model of vehicle to carry out vehicle coupling in the foreground area of extracting.When the type of vehicle, imaging angle and model approach in prospect, can obtain good detection effect, the target outside model be detected to effect not good enough.These class methods need to be take scene geometrical constraint as basis, and the acquisition of these geometric relationships needs larger calculated amount in actual applications, and may not set up; Model Matching is higher to the requirement of the precision of vehicles segmentation and road calibration, and versatility is poor.Method based on characteristic model be utilize hidden target still the feature of visible part mate and realize and block separation.When solving partial occlusion problem, characteristic model the most effectively.When but these class methods need to be supposed to block, selected feature still can be extracted on by screening vehicle.This restriction has caused these class methods to have following defect: can not predict feature used effective verification and measurement ratio when blocking generation, even if adopt the method for many Fusion Features, can not determine in the face of specifically blocking this and select which kind of feature as main criterion.In addition, the selection method due to unique point in these class methods is pre-determined, and for a plurality of similar objects, possible these unique points do not have larger discrimination.Method based on statistical model is also an effective way that solves target occlusion problem, the tracking of shelter target has been obtained to good treatment effect, but parameter estimation and label procedure are very complicated, also very sensitive to the prior model of selecting, the requirement that processing speed does not reach real-time system simultaneously.Method based on inference pattern is to solve occlusion issue according to the priori of the information such as the position of vehicle in scene, track.Many times, all can demonstrate its good performance, yet these class methods are stronger to the dependence of the prior imformations such as the track of vehicle in scene, position, for complicated traffic scene, the method for existing inference pattern is still complete not.
Find by prior art documents, online sample classification is mentioned at some document for blocking the method for processing.As publish an article on < < Pattern recognition > > " Tracking multiple objects through occlusion with online sampling and position estimation " such as L. Zhu, this article utilizes color, texture, the similarity of space feature between successive frame of target, will block segmentation problem and be converted into the classification problem of pixel.Article is pointed out after detecting and blocking, first from blocking the N frame historical data of unshielding target corresponding to prospect with this, by the method for sliding window, extract color, the textural characteristics of pixel in window area, composing training sample, secondly blocking from present frame extracted test sample book after the same method in prospect, finally by the coupling of test sample book and training sample, complete the classification of test sample book, thereby complete, cut apart.Test findings in literary composition shows that the method can overcome the hypothesis of level and smooth movement, still has good matching effect to the target of position flip-flop.Compare to the general dividing method that blocks based on characteristic model and there is high accuracy.But the processing time of the method for this sample online classification is directly proportional to the number of test sample book, is also directly proportional to foreground area.When two vehicle targets have just blocked, containing the foreground target blocking, be almost the summation of these two target areas, therefore, the initial stage of the method between processing target is when block, need the sample size of coupling very huge, can not meet the requirement of Real-time System.Based on prospect profile convex closure, analysis is carried out the method that shielding automobile cuts apart and is mentioned in some document.As publish an article on < < Intelligent Transportation System > > " Multilevel framework to detect and handle vehicle occlusion " such as Zhang Wei, this article utilizes the area of shielding automobile convex closure recently to judge whether to block, and take the point of internal moment maximum and block line of cut as cut point search.The computing of the method is consuming time lower, the initial stage of blocking at vehicle is more effective, but along with the raising of coverage extent between vehicle, the Area Ratio difference of the Area Ratio of shielding automobile convex closure and unshielding vehicle convex closure can be dwindled, and cause blocking, can not effectively be identified and be cut apart.
Therefore, a kind of real-time is good, and accuracy is high, can be applicable to the occlusion handling method of real road environment sensing, to improving the precision of target detection and tracking, there is active influence, for the intelligent level that improves real-time, validity and the traffic administration of transport information, play an important role.
Summary of the invention
For overcoming deficiency of the prior art, the vehicle sections that the present invention aims to provide a kind of two-stage framework blocks removing method, and the method accuracy of detection is high, and real-time is good and applicability is strong.
For realizing above-mentioned technical purpose, the present invention is achieved through the following technical solutions: first utilize the foreground area of blocking and the mutual relationship of blocking between vehicle in front area to carry out the judgement of coverage extent, when
time, for blocking a little less than vehicle, otherwise block by force for vehicle,
be
constantly the
the individual area that blocks prospect;
,
be
constantly with the
individual block prospect corresponding
,
individual vehicle area,
scale factor,
; A little less than adopting again method based on convex closure property analysis to carry out, block and cut apart and adopt the method based on online sample classification to block and cut apart by force.
Compared with prior art, the present invention has following beneficial effect:
The present invention has adopted a kind of occlusion handling method of the two-stage framed structure based on degree of blocking, and level occlusion handling method in the frame based on convex closure analysis is combined with the TRL tracing level occlusion handling method based on online sample classification.Take " degree of blocking " realized the fusion of two kinds of methods as bridge, a little less than making, block the stage, the defect of online sample classification labeling method real-time aspect makes up by the occlusion handling method of cutting apart based on convex closure, make to block the stage by force, convex closure split plot design makes up by the method based on sample online classification in the defect of cutting apart aspect accuracy, realized the mutual supplement with each other's advantages of frame interior level algorithm and TRL tracing level algorithm, reliability and the real-time of partial occlusion separation between vehicle have greatly been improved, for the reliability that improves vehicle detection and tracking, have more greatly and benefit, can be widely used in video monitoring system, in intelligent transportation system and all kinds of civilian system.
Accompanying drawing explanation
Fig. 1 is the process flow diagram that the vehicle sections of a kind of two-stage framework of the present invention blocks removing method;
Fig. 2 is shielding automobile and unshielding vehicle and convex closure schematic diagram separately.Wherein, Fig. 2 (a), Fig. 2 (b) are respectively non-shielding automobile and convex closure schematic diagram thereof; Fig. 2 (c), Fig. 2 (d) are respectively two vehicles that block mutually and convex closure schematic diagram thereof;
Fig. 3 is the schematic diagram of cutting apart of weak shielding automobile, and the convex template that wherein Fig. 3 (a) is shielding automobile is poor; The cut point of Fig. 3 (b) for selecting; Fig. 3 (c) is possible " line of cut " and final " line of cut "; The shielding automobile of Fig. 3 (d) for cutting apart with " line of cut ".
Fig. 4 is cutting apart of strong shielding automobile, and wherein Fig. 4 (a) is for strong shielding automobile prospect occurs; The test sample book center point set that Fig. 4 (b) obtains for sliding window mode; Fig. 4 (c) is for carrying out the result of mark to pixel after sample classification; The vehicle tracking result of Fig. 4 (d) for representing with rectangle frame.
Fig. 5 adopts the vehicle tracking result obtaining after removing method of blocking of the present invention.
Fig. 6 blocks processing time comparing result in the tracing process corresponding with Fig. 5.
Embodiment
Technical scheme for a better understanding of the present invention, is further described embodiments of the present invention below in conjunction with accompanying drawing.
The present invention, to the foreground area of blocking, introduces the concept of degree of blocking.First according to coverage extent, will block a little less than being divided into and block and block by force; Secondly for weak, block, adopt the occlusion handling method that in frame, level is analyzed based on prospect profile convex closure; Finally for blocking by force, adopt the occlusion handling method of TRL tracing level based on sample online classification.In fact, the present invention proposes the concept of degree of blocking, and for bridge, realized two kinds according to this and block the degree of depth of dividing method, effectively merge, utilize advantage separately to make up the other side's defect, reach whole more excellent performance.For the extracted prospect of blocking, according to the group method shown in Fig. 1, block processing.The concrete implementation detail of each several part is as follows:
1, strong and weak shadowing
Vehicle belongs to rigid-object, and the area attribute in operational process in two-dimensional imaging plane has continuity.The present invention according to the area of the prospect of blocking and with this block prospect corresponding do not block time foreground area between relation, the coverage extent that blocks foreground area is divided.A little less than being specifically divided into, block and block by force.The vehicle of rigid body type, its numerous features (as the area on travelling speed, running orbit and two dimensional image) present flatness and gradually changeable in continuous videos sequence.And block, be exactly in original tracked object region, the phenomenon that continous-stable has certain size to be covered by other image-regions occurs in some directions, make original tracked object region area reducing, this process is progressive.Therefore the mutual relationship that can utilize the foreground area area blocking and block between vehicle in front area is carried out the judgement of coverage extent.If in picture frame
individual shielding automobile prospect is by
and the
the vehicle prospect of individual unshielding develops.To meeting blocking of formula (1), a little less than being defined as, the present invention blocks, and do not meet blocking of this formula and be defined as by force and block.
Wherein,
be
constantly the
the individual area that blocks prospect;
with
represented
constantly with the
the individual prospect correspondence of blocking
individual and
the area of individual vehicle region.
be a scale factor, the present invention gets
.
2, in frame level a little less than block processing
For weak, block, the present invention has adopted the dividing method that blocks based on the analysis of prospect profile convex closure.First adopting the Graham(Ge Li permanent) scanning method obtains the convex closure of prospect; Secondly on prospect profile, find cut-point; Finally take cut point as summit, find out the cut-off rule of best Area Ratio, its segmentation result is as blocking separated result.Processing procedure comprises that convex closure extracts, cut-point extracts, line of cut extracts several steps:
1) convex closure extracts
Under normal circumstances, the shape of unshielding vehicle is convex, but can be different by the shape of the vehicle of partial occlusion.The present invention obtains the area of convex closure by Graham scanning method.Fig. 2 provided calculate after shielding automobile and unshielding vehicle and foreground detection convex closure separately.To the non-screening vehicle shown in Fig. 2 (a), the convex closure obtaining by Graham scanning method, as shown in C1 in Fig. 2 (b), can be found out: unshielding vehicle can with the matching very closely of its convex closure; To the shielding automobile shown in Fig. 2 (c), the convex closure obtaining by Graham scanning method, as shown in C2 in Fig. 2 (d), can be found out: between shielding automobile and its convex closure, occurred very large gap.
2) cut-point extracts
The cut-point leaching process that the present invention adopts comprises two steps:
Step 1: as shown in black fill area in Fig. 3 (a), can have a lot of isolated areas between shielding automobile and its convex closure C2, find out wherein maximum isolated area and be designated as
, calculate
convex closure
, establish
for
vertex set,
number for summit;
Step 2: the vertex set of establishing whole shielding automobile profile is
,
number for summit.Can find out set
with set
between must have coincidence, from set
in choose and gather
what do not overlap is any
summit, and calculate distance in it according to formula (2).As shown in Fig. 3 (b),
in edge sequence, there is the point of imperial palace distance
, this point
be chosen as cut point;
3) line of cut extracts
The line of cut leaching process that the present invention adopts is as follows:
With cut point
for summit, choose different inclination angles, make a series of line of cut, as shown in Fig. 3 (c), each line of cut can become foreground segmentation two isolated areas, by the Area Ratio shown in formula (3)
account form, finds the cut-off rule that possesses best Area Ratio.Calculate the Area Ratio sum in these two regions, when this value hour, corresponding line of cut is confirmed as final " line of cut ", and in Fig. 3 (c), ray S1 is exactly the line of cut obtaining through above method, has realized the ideal shown in S2 in Fig. 3 (d) and has cut apart.
Wherein,
for the poor region area of convex,
for the area of the shielding automobile that extracts,
area for shielding automobile convex closure region.
3, TRL tracing level blocks by force processing
For blocking by force, the present invention has adopted the method for sample online classification to block to cut apart.From blocking prospect of present frame, utilize the method for sliding window to extract test sample book, the characteristic quantity extracting from occlusion area is considered as to test sample book; From with this block prospect corresponding do not block time target history image adopt the method for same sliding window to extract training sample, from stored the block characteristic quantity that foreground area generation in extract corresponding with the prospect of blocking, be considered as training sample, different vehicle targets has different specimen number; Employing KNN(K Nearest Neighborhood) (k nearest neighbor) method is found with test sample book and is mated target sample most in training sample, and this training sample adopts the label that mates training sample most to carry out mark.The result of mark is corresponding blocks the result of cutting apart.Therefore blocking separation problem is just converted into the classification problem of utilizing training sample to complete test sample book.
Method based on online sample classification is carried out vehicle and is blocked and cut apart by force, and processing procedure comprises following several step:
1) extract training sample
For each unshielding vehicle target in effective monitoring region in present frame, set up a dynamic data base, store the up-to-date N frame foreground image (N of the present invention gets 20) of its operational process, and carry out mark with tracking label.So, a certain tracking label is
unshielding vehicle, its training image sequence
shown in formula (4).
(4)
Wherein,
for
constantly, following the tracks of label is
unshielding vehicle foreground image.
maximum index value for prospect.When not blocking generation, these dynamic data sequence are upgraded frame by frame, until unshielding vehicle rolls guarded region away from, corresponding storage data are deleted with it; When blocking, training sample just obtains in these unshielding vehicle prospect sequences corresponding with the prospect of blocking.
If
in present frame, prospect label is constantly
mprospect
, be by
with
individual unshielding vehicle blocks rear generation, processes so this label to be
mblock prospect time, its training sample will extract and obtain from the T1 shown in formula (5) and T2 image sequence:
The present invention adopts the mode of sliding window to carry out training sample extraction, is specially:
The square window of one L*L size is carried out along horizontal and vertical direction in foreground area
the traversal of step-length, in extraction window, color, the textural characteristics of pixel, use this vector
characterize the pixel property of window center point.If when in window, the ratio of foreground area and window area is less than 50%, this sample is dropped.L generally gets 25.Thus, can obtain this square window corresponding
dimension training sample vector
, training sample carries out mark with corresponding tracking label, and the All Ranges of same target prospect all has same mark, and different target has different marks.
2) extract test sample book
Test sample book is the foreground area of blocking in present frame
middle extraction, the extracting mode of test sample book is identical with the extracting mode of training sample, and therefore not to repeat here.
3) sample matches
Be specially and adopt KNN method (K=3) to find and mate target sample most with training sample in test sample book, this training sample adopts the label that mates training sample most to carry out mark.The result of mark is corresponding blocks the result of cutting apart.For blocking by force between the vehicle shown in Fig. 4 (a), adopt the test sample book central point of sliding window method extraction as shown in Fig. 4 (b).For in test sample book
individual sample vector
, utilize formula (6) to calculate the distance of itself and training sample, find out three training samples with bee-line
,
,
.
If wherein belong to the sample size of T1, be
ni1, the sample size that belongs to T2 is
ni2.If
ni1>
ni2, this test sample book is marked as the classification of T1
, otherwise be marked as
.
Based on said method, complete the mark of training sample.Adopt subsequently 8 neighborhoods to carry out smoothing processing to the result of mark, to remove isolated noise.The segmentation result corresponding with Fig. 4 (a), obtains Fig. 4 (c) after filling by two kinds of gray scales.Fig. 4 (d) is respectively 1 and 2 vehicle tracking result for the tracking label that represents with rectangle frame, and wherein rectangle frame has represented the extraneous rectangle of minimum of prospect.
According to as above step, Fig. 5 has provided four moments that the tracking label corresponding with Fig. 4 is respectively 1 and 2 shielding automobile tracing process, and wherein, numeral 1 and 2 has represented vehicle tracking label separately, and rectangle frame has represented vehicle minimum boundary rectangle separately.From tracking results, can find out that the present invention suggests plans and can be good at guaranteeing the accurate tracking of the prospect of blocking.For this example, the consuming time contrast of computing of the present invention's computing of blocking processing used is consuming time and document " Tracking multiple objects through occlusion with online sampling and position estimation " institute's extracting method, comparing result as shown in Figure 6, wherein in Fig. 6, the lines of a band circle have represented that the computing of the method for the invention is consuming time above, and in Fig. 6, the leg-of-mutton lines of a band band have represented that the computing of contrast algorithm is consuming time below.Can find out, the present invention can be good at suppressing it and blocks the higher consuming time of period weak, and real-time is guaranteed preferably.And the present invention suggests plans and can solve the strong shielding automobile segmentation problem that convex closure split plot design cannot be tackled.
Claims (4)
1. the vehicle sections of two-stage framework blocks a removing method, it is characterized in that: first utilize the foreground area of blocking and the mutual relationship of blocking between vehicle in front area to carry out the judgement of coverage extent, when
time, for blocking a little less than vehicle, otherwise block by force for vehicle,
be
constantly the
the individual area that blocks prospect;
,
be
constantly with the
individual block prospect corresponding
,
individual vehicle area,
scale factor,
; A little less than adopting again method based on convex closure property analysis to carry out, block and cut apart and adopt the method based on online sample classification to block and cut apart by force.
2. a kind of vehicle sections of two-stage framework blocks removing method according to claim 1, it is characterized in that: the method for described employing based on convex closure property analysis blocked the step of cutting apart a little less than carrying out and be:
1) convex closure extracts, and by Graham scanning method, obtains the convex closure C1 of non-screening vehicle and the convex closure C2 of shielding automobile,
2) cut-point extracts, and finds out the maximum isolated area existing between shielding automobile and its convex closure C2
, calculate maximum isolated area
convex closure
, from convex closure
vertex set in choose any summit not overlapping with the vertex set of whole shielding automobile profile
, calculate distance in it, at convex closure
in edge sequence, choosing has the summit of imperial palace distance
for cut point;
3) line of cut extracts, with cut point
for summit, choose different inclination angles and make a series of line of cut, each line of cut becomes two isolated areas by foreground segmentation, calculates the Area Ratio sum of these two isolated areas, when this Area Ratio sum hour, corresponding line of cut is final line of cut.
3. a kind of vehicle sections of two-stage framework blocks removing method according to claim 1, it is characterized in that: the method for described employing based on online sample classification blocked by force the step of cutting apart and be:
1) extract test sample book and training sample, from occlusion area, extract characteristic quantity as test sample book, from stored corresponding with the prospect of the blocking characteristic quantity that extracts in the foreground area generation of blocking as training sample;
2) sample matches, in test sample book, find and mate target sample most with training sample, this training sample adopts the label that mates training sample most to carry out mark, adopts 8 neighborhoods to carry out smoothing processing to the result of mark, and the result correspondence of mark is blocked the result of cutting apart.
4. a kind of vehicle sections of two-stage framework blocks removing method according to claim 3, it is characterized in that: the method for extracting training sample is: for unshielding vehicle target, set up a dynamic data base, store the up-to-date N frame foreground image of its operational process, and carry out mark with tracking label, when not blocking generation, these dynamic data sequence are upgraded frame by frame, when vehicle rolls away from after guarded region, corresponding storage data are deleted with it, when blocking, training sample just obtains in these unshielding vehicle prospect sequences corresponding with the prospect of blocking; The extracting mode of test sample book is identical with the extracting mode of training sample.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310674807.9A CN103632376A (en) | 2013-12-12 | 2013-12-12 | Method for suppressing partial occlusion of vehicles by aid of double-level frames |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310674807.9A CN103632376A (en) | 2013-12-12 | 2013-12-12 | Method for suppressing partial occlusion of vehicles by aid of double-level frames |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103632376A true CN103632376A (en) | 2014-03-12 |
Family
ID=50213393
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310674807.9A Pending CN103632376A (en) | 2013-12-12 | 2013-12-12 | Method for suppressing partial occlusion of vehicles by aid of double-level frames |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103632376A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105894542A (en) * | 2016-04-26 | 2016-08-24 | 深圳大学 | Online target tracking method and apparatus |
CN106991684A (en) * | 2017-03-15 | 2017-07-28 | 上海信昊信息科技有限公司 | Foreground extracting method and device |
CN107341490A (en) * | 2017-07-17 | 2017-11-10 | 华中科技大学 | A kind of shielding automobile detection method and system based on convex closure analysis |
CN110009929A (en) * | 2019-03-15 | 2019-07-12 | 北京筑梦园科技有限公司 | A kind of Vehicle berth management method, equipment and system |
CN111191309A (en) * | 2019-12-19 | 2020-05-22 | 北京虹晟信息科技有限公司 | Real-time shielding and rejecting method, device and equipment suitable for BIM light weight |
CN112116634A (en) * | 2020-07-30 | 2020-12-22 | 西安交通大学 | Multi-target tracking method of semi-online machine |
CN113657462A (en) * | 2021-07-28 | 2021-11-16 | 讯飞智元信息科技有限公司 | Method for training vehicle recognition model, vehicle recognition method and computing device |
CN114140501A (en) * | 2022-01-30 | 2022-03-04 | 南昌工程学院 | Target tracking method and device and readable storage medium |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103310465A (en) * | 2013-06-27 | 2013-09-18 | 东南大学 | Vehicle occlusion treating method based on Markov random field |
-
2013
- 2013-12-12 CN CN201310674807.9A patent/CN103632376A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103310465A (en) * | 2013-06-27 | 2013-09-18 | 东南大学 | Vehicle occlusion treating method based on Markov random field |
Non-Patent Citations (3)
Title |
---|
LIN ZHU ET AL.: "Tracking multiple objects through occlusion with online sampling and position estimation", 《PATTERN RECOGNITION》 * |
WEI ZHANG ET AL.: "Multilevel Framework to Detect and Handle Vehicle Occlusion", 《IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTTATION SYSTEMS》 * |
杨华: "基于视频的城市道路车辆自动检测与跟踪", 《国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105894542A (en) * | 2016-04-26 | 2016-08-24 | 深圳大学 | Online target tracking method and apparatus |
CN105894542B (en) * | 2016-04-26 | 2019-06-11 | 深圳大学 | A kind of online method for tracking target and device |
CN106991684A (en) * | 2017-03-15 | 2017-07-28 | 上海信昊信息科技有限公司 | Foreground extracting method and device |
CN106991684B (en) * | 2017-03-15 | 2019-09-17 | 上海信昊信息科技有限公司 | Foreground extracting method and device |
CN107341490A (en) * | 2017-07-17 | 2017-11-10 | 华中科技大学 | A kind of shielding automobile detection method and system based on convex closure analysis |
CN110009929A (en) * | 2019-03-15 | 2019-07-12 | 北京筑梦园科技有限公司 | A kind of Vehicle berth management method, equipment and system |
CN111191309A (en) * | 2019-12-19 | 2020-05-22 | 北京虹晟信息科技有限公司 | Real-time shielding and rejecting method, device and equipment suitable for BIM light weight |
CN111191309B (en) * | 2019-12-19 | 2023-01-24 | 北京虹晟信息科技有限公司 | Real-time shielding and rejecting method, device and equipment suitable for BIM light weight |
CN112116634A (en) * | 2020-07-30 | 2020-12-22 | 西安交通大学 | Multi-target tracking method of semi-online machine |
CN113657462A (en) * | 2021-07-28 | 2021-11-16 | 讯飞智元信息科技有限公司 | Method for training vehicle recognition model, vehicle recognition method and computing device |
CN114140501A (en) * | 2022-01-30 | 2022-03-04 | 南昌工程学院 | Target tracking method and device and readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103632376A (en) | Method for suppressing partial occlusion of vehicles by aid of double-level frames | |
Luvizon et al. | A video-based system for vehicle speed measurement in urban roadways | |
Geiger et al. | Are we ready for autonomous driving? the kitti vision benchmark suite | |
Yamaguchi et al. | Vehicle ego-motion estimation and moving object detection using a monocular camera | |
CN103324913B (en) | A kind of pedestrian event detection method of Shape-based interpolation characteristic sum trajectory analysis | |
Zhou et al. | A fast and accurate segmentation method for ordered LiDAR point cloud of large-scale scenes | |
Broggi et al. | Self-calibration of a stereo vision system for automotive applications | |
Chen et al. | Next generation map making: Geo-referenced ground-level LIDAR point clouds for automatic retro-reflective road feature extraction | |
CN104978567B (en) | Vehicle checking method based on scene classification | |
CN104850850A (en) | Binocular stereoscopic vision image feature extraction method combining shape and color | |
Li et al. | Automatic registration of panoramic image sequence and mobile laser scanning data using semantic features | |
CN104517095B (en) | A kind of number of people dividing method based on depth image | |
CN104700414A (en) | Rapid distance-measuring method for pedestrian on road ahead on the basis of on-board binocular camera | |
Soheilian et al. | 3D road marking reconstruction from street-level calibrated stereo pairs | |
CN104063882B (en) | Vehicle video speed measuring method based on binocular camera | |
CN103206957B (en) | The lane detection and tracking method of vehicular autonomous navigation | |
CN102243705B (en) | Method for positioning license plate based on edge detection | |
CN110197173B (en) | Road edge detection method based on binocular vision | |
Ye et al. | Robust lane extraction from MLS point clouds towards HD maps especially in curve road | |
CN104851089A (en) | Static scene foreground segmentation method and device based on three-dimensional light field | |
CN106156752A (en) | A kind of model recognizing method based on inverse projection three-view diagram | |
CN108416798B (en) | A kind of vehicle distances estimation method based on light stream | |
CN112070756A (en) | Three-dimensional road surface disease measuring method based on unmanned aerial vehicle oblique photography | |
Sebsadji et al. | Robust road marking extraction in urban environments using stereo images | |
Schlichting et al. | Vehicle localization by lidar point correlation improved by change detection |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20140312 |