CN106340031A - Method and device for detecting moving object - Google Patents
Method and device for detecting moving object Download PDFInfo
- Publication number
- CN106340031A CN106340031A CN201511001426.XA CN201511001426A CN106340031A CN 106340031 A CN106340031 A CN 106340031A CN 201511001426 A CN201511001426 A CN 201511001426A CN 106340031 A CN106340031 A CN 106340031A
- Authority
- CN
- China
- Prior art keywords
- region
- image
- road surface
- moving object
- carried out
- 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
- 238000000034 method Methods 0.000 title claims abstract description 81
- 238000001514 detection method Methods 0.000 claims abstract description 45
- 230000033001 locomotion Effects 0.000 claims abstract description 35
- 239000013598 vector Substances 0.000 claims abstract description 31
- 230000008569 process Effects 0.000 claims abstract description 11
- 238000001914 filtration Methods 0.000 claims description 13
- 230000001133 acceleration Effects 0.000 claims description 3
- 230000003287 optical effect Effects 0.000 abstract description 2
- 238000007670 refining Methods 0.000 abstract 1
- 238000004364 calculation method Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000006073 displacement reaction Methods 0.000 description 2
- 238000009394 selective breeding Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004040 coloring Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 231100001261 hazardous Toxicity 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
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
-
- 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
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30248—Vehicle exterior or interior
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
A method and apparatus for detecting a moving object. The method of the invention carries out moving object detection by reusing the motion vector generated in video coding, and can detect the moving object with larger relative speed with the vehicle in real time. The method adopts the idea of image blocks in video coding, replaces optical flow with motion vectors generated in the video coding process, and uses space constraint conditions based on the image blocks to detect moving objects. Before the moving object detection is carried out, the road surface identification is carried out on the image, and the area for carrying out the moving object detection subsequently is reduced. Based on the space constraint method of the image blocks, the image blocks which do not meet the space constraint condition are marked as motion image blocks, then the motion image blocks are screened and clustered to obtain an initial motion object. And finally, refining the initial moving object to obtain a final detection result.
Description
Technical field
The present invention relates to the method and apparatus of detection moving object, specifically, relate to the use of video
The method and apparatus to detect moving vehicle for the motion vector in coding.
Background technology
Vehicle testing techniques are critically important links in intelligent transportation.With automobile pollution
Increase, in the face of the traffic environment becoming increasingly complex, occur the probability of vehicle accident to be substantially increased.Hand over
Interpreter thus cause great loss to the country and people.In order to reduce the generation of tragedy, increasingly
Many manpower and materials have been put in the research of car steering aid system.By by some science and technology
Means, the traffic environment residing for perception driver, and information is transmitted in the form of video or audio frequency
To driver, him is helped to make correct judgement, the generation of minimizing accident.
Vehicle testing techniques are always the study hotspot of intelligent transportation field, also achieve in recent years not
Few achievement.
Vehicle detection can be carried out by radar, laser radar and machine vision etc..Based on radar or
It is the method for laser radar, by sending a kind of continuous FM signal, signal reaches moving object
It is reflected back after thing, transmitting terminal is demodulated and frequency analyses to the signal receiving, and finds fortune
Animal body.The method of view-based access control model mainly carries out image acquisition: monocular shooting using two kinds of photographic head
Head and binocular camera.Binocular camera can reflect the depth information of scene, but Installation And Calibration ratio
Cumbersome, and price is also relatively expensive.Monocular cam low price, but scene can not be embodied
Depth information.The vehicle checking method of view-based access control model is broadly divided into two big class: based on surface
With based drive method.Method based on surface mainly utilizes the resemblance of vehicle, such as
Symmetry, straight line, shade etc. naturally it is also possible to using statistical learning method, extract some normal
The feature seen, such as hog, the feature such as haar, gabor, recycle the method for machine learning to carry out
Vehicle detection.This kind of calculation speed is fast, can meet the requirement of real-time, but because vehicle is being moved through
There may be in journey and block, scene light mutabilityization, different visual angles produce different shape etc., this
Class method does not often have robustness.Need to gather substantial amounts of sample based on the method for statistical learning
Row training, and to different types of moving object, need to train different graders, this kind of side
Method is more complicated.Based drive vehicle checking method, mainly uses the letter such as light stream, disparity map
Cease and to be detected, the advantage of this kind of method is to can be suitably used for several scenes, but calculate too complicated,
Need good hardware resource, method can be only achieved real-time.
The main technical problem to be solved in the present invention is to provide the base being available for drive assist system use
In the Mobile object detection method of machine vision, moving object can be not limited to automobile, be also not necessarily limited to it
Visual angle.The method utilizes motion vector in Video coding, has real-time, and has certain
Robustness, and the function that is provided simultaneously with making video recording in real time.
The present invention utilizes motion vector to replace light stream to be used for moving object segmentation.Calculating due to light stream
It is a very time-consuming process, particularly calculate dense optical flow, therefore replace light stream with motion vector
Afterwards, both save resources, save the time again, have ensured the real-time of method.
The thought of piecemeal in Video coding is applied in moving object segmentation the present invention.Based on pixel
Point move this object computationally intensive, be extremely difficult to real-time, due to image spatially have similar
Property, to non-edge, the gray value of pixel in pixel and its field and color data error are not
Can be too big, carrying out moving object segmentation therefore in units of segment can't affect Detection results, with
When can reduce amount of calculation.
Content of the invention
The invention provides a kind of method of detection moving object, comprising:
(1) shoot the sequence of pavement image, and produce motion vector information；
(2) pavement detection is carried out to image, identify road surface region；
(3) image is carried out with piecemeal, removes road surface region, obtain non-road surface region；
(4) according to motion vector information, identify the moving block in non-road surface region；
(5) relatively slow speed and relatively rapid region are sub-divided into non-road surface region；
(6) moving block in relatively rapid region is carried out filtering, clusters (cluster), obtain relatively
Fast moving objects；
(7) image in relatively slow speed region is carried out filtering, clusters, obtain target moving object.
Preferably, in the above-mentioned methods, described pavement detection uses with segment as ultimate unit
Region growing method.
Preferably, in the above-mentioned methods, described pavement detection uses the meansigma methodss of color histogram
With standard deviation as the main information in area growth process.
Preferably, in the above-mentioned methods, described pavement detection is only carried out to the latter half of image.
Preferably, in the above-mentioned methods, wherein said relatively slow speed is sub-divided into non-road surface region
And relatively rapid region is that size according to motion vector is carried out.
Preferably, in the above-mentioned methods, the wherein said moving block to relatively rapid region is carried out
Filtration is carried out using the constraint of size, position and direction.
Preferably, in the above-mentioned methods, the wherein said moving block to relatively rapid region is carried out
Cluster after filtration, method be by the size of plane parallax residual vector ppmv, direction and away from
Compare from specific threshold value.
Preferably, in the above-mentioned methods, the wherein said moving block to relatively slow speed region is carried out
Cluster after filtration, its method is to find out road surface gray value the darkest during through pavement detection
Then form bianry image with this gray value, then obtain horizontal and vertical gradient map with sobel filter
After picture, position the position of relative low speeds object with bianry image and horizontal gradient image.
Preferably, in the above-mentioned methods, wherein become in the moving object detection to relatively slow speed region
After work(, enter target following pattern.
Present invention also offers a kind of device for detecting moving object, comprising:
Video camera, is configured to shooting image sequence；
Processor, is configured to carry out pavement detection to image, identifies road surface region；Image is entered
Row piecemeal, removes road surface region, obtains non-road surface region；According to motion vector information, identify
Go out the moving block in non-road surface region；Moving block is filtered, clustered and is refined, transported
Animal body.
Preferably, said apparatus also include display, for display image sequence, and indicate detection
The target moving object arrived.
Preferably, said apparatus also include sensor, for obtaining car during advancing for the vehicle
The dynamic parameter of speed, acceleration and video camera.
Preferably, the dynamic parameter of above-mentioned video camera is the angle of pitch of video camera, yaw angle.
Preferably, said apparatus also include the interface that external device (ED) is connected, for detection message
Externally send, receive exterior vehicle signal such as speed, rain brush (wiper), brake signal, or
Receive external command such as functional switch.
Brief description
Fig. 1 is based drive moving vehicle detection method schematic diagram.
Fig. 2 shows the process in relatively slow speed region to non-road surface region and relatively rapid region, its
Middle Fig. 2-1 shows the typical image (Fig. 2-1a) being captured by camera, and the converted ash obtaining
Degree image (Fig. 2-1b)；Fig. 2-2 and Fig. 2-3 shows the process to relatively slow speed region；Fig. 2-4
Show the process to relatively rapid region.
Fig. 3 is car tracing flow chart.
Fig. 4 shows the tracking to target vehicle.
Specific embodiment
The present invention is described in detail below in conjunction with the accompanying drawings.
The method of the detection moving object that the present invention provides can use the flowchart representation shown in Fig. 1,
The method mainly comprises the steps that
One. obtain motion vector information
The present invention uses the monocular cam that resolution is 640 × 480 and (can also adopt other
Resolution).Photographic head is arranged on the windshield of automobile, in order to obtain accurate motion compensation knot
Really, photographic head is also installed six axle sensors, for obtaining photographic head during advancing
Acceleration and other related dynamic parameters, the such as angle of pitch of video camera, yaw angle etc..Speed is then sharp
The velocity sensor being carried with automobile, or obtain through communication interface.
The video that photographic head collects is encoded, the motion vector producing in storage cataloged procedure
Information (Fig. 2).
The present invention is using h.264/avc video standard or other video coding techniques based on segment
(such as hevc), can select the size of suitable segment according to the complexity of image.?
In each motion vector, the size of segment is preferably consistent.The frame per second of video is 30fps or more, i.e. two frame
Between time interval be 33ms or following.
In Video coding, in order to reduce code check, mainly employ frame in and inter-frame forecast mode,
When only adopting inter prediction, just can produce motion vector.H.264/avc video standard allows often
Individual captured image is encoded as i frame, p frame, b frame.In the present invention, in order to ensure p frame
Each of segment have motion vector, setting coding mode be inter prediction.
The displacement of vehicle can be compensated by known method, for example advanced driver's auxiliary system
Method used by system (advanced driver assistance systems, abbreviation adas).
Two. pavement detection
Pavement detection is carried out to input picture, obtains secondary road surface figure (Fig. 3).
Because road surface is continually changing, the difficulty of pavement detection is very big, needs to process road surface appearance difference
Multiple situations of object are it is also desirable to avoid the different lighting conditions due to shade and reflective formation as far as possible
Under cause road surface identification mistake.Additionally, it must be efficient that method for distinguishing is known on road surface, so that real
Shi Jinhang.
The present invention uses the region growing method with segment as ultimate unit.Under normal circumstances,
Road surface only appears in the latter half of image, so when actually carrying out region growing, in image
The latter half carry out.In area growth process, use the color histogram of segment
Average and covariance information, rather than simple half-tone information or colouring information, alleviate illumination variation
Etc. the impact to testing result for the factor.
Pavement detection uses the new region growing method based on segment.Select image base
Suitable segment is as seed.By the meansigma methodss between relatively adjacent segment and standard deviation, make base
Grow around seed segment in the region of segment.When more segment cannot be generated, stop road surface
Detection.Specifically comprise the following steps that
1. set up road surface color model
Several segments, preparation road surface color model are chosen on the road surface from multiple images.With public affairs
Formula (7.1) and (7.2) are calculating the meansigma methodss of each segmentWith standard deviation (σ), wherein (x, y) is segment
Origin coordinates, r (u, v), g (u, v), b (u, v) are the red, green, blue chrominance channels of the pixel at (u, v) place
Intensity, hist (i) is the number of the average of three Color Channels and the pixel being equal to i.Artificial selection
One group of different image of light intensity on different road surfaces, the segment on road surface in each image
It is the artificial selection of bottom section from the image against vehicle front.Must not select with road marking
The segment of will.
Obtain about 2500 samples altogether from one group of image, setting up one willWith certain limit
σ be associated form.When initially selecting the road surface for region growing, with this form conduct
Color model.To often a photographic head, this step is only carried out once.The color model meeting being obtained
It is stored in system and use.
2. the seed segment of road surface region growing
After capturing real time imaging, assessment is near the segment of image baseAnd σ.This be because
For when the vehicle is running, this region is the region most unlikely with mobile object.Some segments can
Can be able to be affected by pavement marker.IfOr the numerical value that σ limits beyond color model, then
Select another segment along same a line (row) again, to assess againAnd σ, until new segment
Meet the requirement of color model with σ.
3. road surface region growing
Select, in image base a line, the segment that size is 16 × 16, store the average of this segment afterwards
ValueNumerical value with standard deviation (σ).Because the surface on road generally has uniform color and weak texture,
Some segment in road surface regionThe segment should being adjacent with σClose with σ.
When near image base a line segment in identifying with meansigma methodssAnd standard deviation sigmai's
Segment biAfter being road surface region segment, also can calculate and segment biEight adjacent segments average
ValueAnd standard deviation sigmak, wherein k represents one of eight segments.Below meeting (7.3) and (7.4)
During formula, adjacent segment bkIt is identified as road surface region segment, b and c in formula is to preset
Threshold value:
||ak-ai| | ＜ b (7.3)
||σk-σi| | ＜ c (7.4)
By comparing the road surface region segment identifying and the segment adjacent with them, area can be adopted
Domain growing method is making road surface region further " growth ".
4. the improvement after the region growing of road surface
Due to when being compared,Be unsatisfactory for (7.3) and (7.4) formula with the numerical value of σ, road surface some
Segment can be excluded by road surface region growing method, occurs " empty ".It is thus desirable to post processing is filling
" empty ", thus improve detected road surface region.
Three. obtain non-road surface region
Input is divided into a certain size segment, combining road figure, obtains needs and subsequently located
Manage non-road surface region, thus needing image-region to be processed when reducing detection moving object.
Four. mark moving block
Near extension focus (focus of expansion, foe, see Fig. 2-1 (b)) motion vector with
And the motion vector of relatively slow speed mobile object may be very little.Due to h.264/avc encoder
Limited precision, little motion vector is inaccurate.Accordingly, it would be desirable to additive method is relatively slow to detect
The mobile object of speed.
Size (amplitude) according to motion vector (mv), non-road surface region is divided into relatively slow speed
Region and relatively rapid region.
1. the target area (relatively slow speed region) of relatively slow speed object
Motion vector due to having the region of relatively slow speed mobile object is less, therefore can select
The less region of motion vector is as the target area of relatively slow speed object.Additionally, in order to simplify place
Reason program, can reduce target area by detecting road surface region further it is also possible to by target area
Domain is limited to foe area below.
Fig. 2-1 (a) shows the typical image being captured by camera, has been converted into gray level image.Due to
The top of the focus (foe) of extension is typically the top of sky or mobile vehicle, therefore can be ignored
Fig. 2-1 (b).
Fig. 2-2 (a) shows the road surface area being detected with the method for " pavement detection " above part description
Domain figure cover (mask).Fig. 2-2 (b) shows that motion vector is more than threshold value qmFigure cover, therein white
Zone domain is that motion vector is more than qmRegion.Fig. 2-2 (c) be the figure cover of (a) and (b) is combined after,
Remove the target area that foe above section obtains again.When carrying out relatively slow speed moving Object Detection,
Do not consider white segment therein.
Still some can remove for the target area obtaining in aforementioned manners, such as Fig. 2-3 (a)
Shown in middle encircled portion.The method removing is that the maximum y of every string white print block in record image sits
Mark and minimum y-coordinate.If certain segment is located between minimum and maximum coordinate in certain string,
This segment is labeled as white, represents that this segment can be ignored.The target area so processing is such as
Shown in Fig. 2-3 (b).
Obviously, the region for detecting relatively slow speed vehicle obtaining in aforementioned manners is far smaller than former
Beginning image, such that it is able to significant shortening detection time.
2. the target area (relatively rapid region) of relatively rapid object
Detect the method for target area and the above detection relatively slow speed object of relatively rapid object
The method of target area is similar to, include pavement detection (Fig. 2-4 (a)) and making motion vector more than or
Equal to threshold value qmFigure cover (Fig. 2-4 (b)), obtain (Fig. 2-4 (c)) after the two is combined.
Five. identification target moving vehicle
1. the moving block in pair relatively rapid region filters, and is then clustered
(1) filter
Filtration is carried out using the constraint of size, position and direction.
What constraint considered is the relation between adjacent two frames.If the screen of present frame point is sat
It is designated as (x2, y2), this corresponding coordinate in former frame is (x1g, y1g), then the remaining arrow of plane parallax
Amount (ppmv) (μx, μy) can be expressed as:
μx=x1g-x1(0.1)
μy=y1g-y1(0.2)
Constraints 1: size
Remaining using plane parallax；Wherein coordinate is the size of the plane parallax remnants of the macro block of (x, y)
Exceed specific threshold μmin(x, y).
Constraints 2: position
A warning area (alert zone) can be set in the picture, then after elapsed time t if
Ppmv enters this area, then related object is considered to enter hazardous area, its ppmv accept into
One step is processed.
Constraints 3: direction
Object and vehicle that ppmv points to the focus (foe) of extension move in parallel itself, or belong to quiet
Only object.They are probably the vehicle of the relatively slow speed of traveling parallel with vehicle itself, or road
Stationary object on face.In any case, the ppmv pointing to foe should all be excluded.This is because
Target area (roi) has eliminated the object of the less ppmv of relatively slow speed vehicle generation, and these are relatively
Little ppmv probably belongs to stationary object.
If the slope of the slope of ppmv and the point pointing to foe is less than threshold value mthres, then can arrange
Ppmv except point (x, y) place.
(2) cluster
By the size of ppmv, direction and distance are compared with specific threshold value, can be poly- to it
Class.
The cluster identifying in aforementioned manners forms module.Module as searching class in new frame.
The original position of search best match is that the size and Orientation according to average motion vector determines.Institute
Storage module and search window from last frame in candidate block between absolute difference it
(sum of absolute difference, sad) represents potential and mates.If absolute difference sum is
In predetermined threshold value, and find local minimum in the search window then it represents that module is successful
Ground coupling.When a match is found then it is assumed that mobile object is detected.Storage former frame with current
The reference of the original position as the search window of next frame for the displacement of module between coupling frame.
2. the moving block in pair relatively slow speed region filters, and is then clustered
Using constraints mentioned above, the moving block in region at a slow speed is filtered, Ran Houjin
Row cluster.
The method that the moving block in relatively slow speed region is clustered is to find out during through pavement detection
Road surface gray value the darkestThen form bianry image with this gray value, filtered with sobel
After device (filter) obtains horizontal and vertical gradient image, positioned with bianry image and horizontal gradient image
The position of relative low speeds object.
3. follow the tracks of target
After the moving object detection success to relatively slow speed region, enter target following pattern.
Because the movement of vehicle to be measured is relatively slow, their size and location when being separated by some frames
Do not have very big change, therefore, in order to reduce amount of calculation, several frames can be skipped in detection, and
Do not interfere with testing result.
Additionally, in order to reduce amount of calculation, can also adopt the present invention tracing algorithm.Due to treating
The original position of measuring car can be learnt from detection process, therefore can be in fair-sized when following the trail of
Search window in check the feature that the position of some vehicles relatively to be measured does not change.This algorithm
Flow process see Fig. 3.
Crossed by (left, top) to (right, bottom) by detecting that relatively slow speed vehicle can identify
Rectangular screen in vehicle, capture a new image, be converted into gray level image.For
The sobel kernel (kernels) finding horizontal and vertical gradient is as follows:
Wherein (a) is look for the sobel kernel of horizontal gradient, and (b) is look for vertical gradient
Sobel kernel.Horizontal and vertical gradient image can be generated according to above-mentioned sobel kernel.Due to
Only using the region near the rectangle crossing during tracking, sobel kernel is only applicable to (a left side -2ex, top
-2ey) to (right side+2ex, bottom+2ey) within region, wherein exAnd eyIt is to sit along screen x respectively
Mark and the pixel count of y-coordinate extension.
Subsequently, calculate (left, bottom-e with vertical gradient imagey) to (right, bottom+ey) in matrix
Floor projection.S (y) in formula (9.4) is the floor projection value of each y-coordinate in specific border:
The gradient of each y-coordinate in border can be found with formula (9.5).S relatively in bordergY () is permissible
Find greatest gradient sgmax(y).
sg(y)=s (y+1)+s (y+2)-s (y-1)-s (y-2) (9.5)
Similarly, with horizontal gradient image in (a left side-ex, (top+bottom)/2) and to (a left side+ex, bottom)
The horizontal outline projection in left side is calculated in rectangle.
S in formula (9.6)gX () is the vertically profiling projection value of each x coordinate in specific border.
The direction that this formula is substantially along y-axis numerical value reduces vertically adds up intensity.So have
The impact discontinuously bringing beneficial to minimizing vertical direction.Maximum sgmax(x) can simply by than
Compared with sgThe value of (x) and obtain.
The horizontal profile projection-type on right side is similar to left side.Difference is for border to be changed to the (right side-ex, (top
+ bottom)/2) to (right side+ex, bottom).
By sgmax(x) and sgmaxY () is compared with predetermined threshold value, if they are in allowed limits,
Then show to the tracking of vehicle it is successful.Subsequently, update the rectangle that crosses, in next frame,
Thus realizing continuous car tracing (Fig. 4).
Above content is further description made for the present invention with reference to specific embodiment,
It cannot be assumed that the present invention be embodied as be confined to these explanations.The affiliated technology neck for the present invention
For the those of ordinary skill in domain, without departing from the inventive concept of the premise, if can also make
Do simple deduction or replace, all should be considered as belonging to protection scope of the present invention.
Claims (14)
1. a kind of method of detection moving object, comprising:
(1) shoot the sequence of pavement image, and produce motion vector information；
(2) pavement detection is carried out to image, identify road surface region；
(3) image is carried out with piecemeal, removes road surface region, obtain non-road surface region；
(4) according to motion vector information, identify the moving block in non-road surface region；
(5) relatively slow speed and relatively rapid region are sub-divided into non-road surface region；
(6) moving block in relatively rapid region is carried out filtering, clusters, obtain relatively rapid motion
Object；
(7) image in relatively slow speed region is carried out filtering, clusters, obtain target moving object.
2. it is substantially single for the method for claim 1 wherein that described pavement detection uses with segment
The region growing method of position.
3. the method for claim 1 wherein that described pavement detection uses the flat of color histogram
Average and standard deviation are as the main information in area growth process.
4. the only the latter half to image of the pavement detection described in the method for claim 1 wherein
Carry out.
5. the method for claim 1 wherein and described relatively slow speed is sub-divided into non-road surface region
And relatively rapid region is that size according to motion vector is carried out.
6. the method for claim 1 wherein that the described moving block to relatively rapid region is carried out
Filtration is carried out using the constraint of size, position and direction.
7. the method for claim 1 wherein that the described moving block to relatively rapid region is carried out
Cluster after filtration, method be by the size of plane parallax residual vector ppmv, direction and away from
Compare from specific threshold value.
8. the method for claim 1 wherein that the described moving block to relatively slow speed region is carried out
Cluster after filtration, its method is to find out road surface gray value the darkest during through pavement detection
Then form bianry image with this gray value, then obtain horizontal and vertical gradient map with sobel filter
After picture, position the position of relative low speeds object with bianry image and horizontal gradient image.
9. the method for claim 1 wherein and become in the moving object detection to relatively slow speed region
After work(, enter target following pattern.
10. a kind of device for detecting moving object, comprising:
Video camera, is configured to shooting image sequence；
Processor, is configured to carry out pavement detection to image, identifies road surface region；Image is entered
Row piecemeal, removes road surface region, obtains non-road surface region；According to motion vector information, identify
The moving block in non-road surface region；Relatively slow speed and relatively rapid area are sub-divided into non-road surface region
Domain；The moving block in relatively rapid region is carried out filtering, clusters, obtains relatively rapid moving object；
The image in relatively slow speed region is carried out filtering, clusters, obtains target moving object.
The device of 11. claim 10, it also includes display, for display image sequence,
And indicate the target moving object detecting.
The device of 12. claim 10, it also includes sensor, is used for obtaining vehicle in traveling
During speed, acceleration and video camera dynamic parameter.
The device of 13. claim 12, the dynamic parameter of wherein said video camera is video camera
The angle of pitch, yaw angle.
The device of 14. claim 10, it also includes the interface that external device (ED) is connected.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| HK15106442.4A HK1203289A2 (en) | 2015-07-07 | 2015-07-07 | A method and a device for detecting moving object |
| HK15106442.4 | 2015-07-07 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106340031A true CN106340031A (en) | 2017-01-18 |
Family
ID=54326703
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201511001426.XA Pending CN106340031A (en) | 2015-07-07 | 2015-12-28 | Method and device for detecting moving object |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN106340031A (en) |
| HK (1) | HK1203289A2 (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106713920A (en) * | 2017-02-22 | 2017-05-24 | 珠海全志科技股份有限公司 | Mobile detection method and device based on video encoder |
| CN107404653A (en) * | 2017-05-23 | 2017-11-28 | 南京邮电大学 | A kind of Parking quick determination method of HEVC code streams |
| CN109426807A (en) * | 2017-08-22 | 2019-03-05 | 罗伯特·博世有限公司 | Method and apparatus for estimating the displacement of vehicle |
| CN110832568A (en) * | 2017-07-05 | 2020-02-21 | 歌乐株式会社 | Vehicle environment recognition device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1960491A (en) * | 2006-09-21 | 2007-05-09 | 上海大学 | Real time method for segmenting motion object based on H.264 compression domain |
| CN101231755A (en) * | 2007-01-25 | 2008-07-30 | 上海遥薇实业有限公司 | Moving target tracking and quantity statistics method |
| CN101494757A (en) * | 2009-01-23 | 2009-07-29 | 上海广电（集团）有限公司中央研究院 | Motion estimating method based on time-space domain mixing information |
| CN101853510A (en) * | 2010-04-20 | 2010-10-06 | 上海大学 | Movement perception model extraction method based on time-space domain |
| CN103733607A (en) * | 2011-08-10 | 2014-04-16 | 富士胶片株式会社 | Device and method for detecting moving objects |
-
2015
- 2015-07-07 HK HK15106442.4A patent/HK1203289A2/en not_active IP Right Cessation
- 2015-12-28 CN CN201511001426.XA patent/CN106340031A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1960491A (en) * | 2006-09-21 | 2007-05-09 | 上海大学 | Real time method for segmenting motion object based on H.264 compression domain |
| CN101231755A (en) * | 2007-01-25 | 2008-07-30 | 上海遥薇实业有限公司 | Moving target tracking and quantity statistics method |
| CN101494757A (en) * | 2009-01-23 | 2009-07-29 | 上海广电（集团）有限公司中央研究院 | Motion estimating method based on time-space domain mixing information |
| CN101853510A (en) * | 2010-04-20 | 2010-10-06 | 上海大学 | Movement perception model extraction method based on time-space domain |
| CN103733607A (en) * | 2011-08-10 | 2014-04-16 | 富士胶片株式会社 | Device and method for detecting moving objects |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106713920A (en) * | 2017-02-22 | 2017-05-24 | 珠海全志科技股份有限公司 | Mobile detection method and device based on video encoder |
| CN107404653A (en) * | 2017-05-23 | 2017-11-28 | 南京邮电大学 | A kind of Parking quick determination method of HEVC code streams |
| CN107404653B (en) * | 2017-05-23 | 2019-10-18 | 南京邮电大学 | A kind of Parking rapid detection method of HEVC code stream |
| CN110832568A (en) * | 2017-07-05 | 2020-02-21 | 歌乐株式会社 | Vehicle environment recognition device |
| CN110832568B (en) * | 2017-07-05 | 2022-03-25 | 歌乐株式会社 | Vehicle environment recognition device |
| CN109426807A (en) * | 2017-08-22 | 2019-03-05 | 罗伯特·博世有限公司 | Method and apparatus for estimating the displacement of vehicle |
Also Published As
| Publication number | Publication date |
|---|---|
| HK1203289A2 (en) | 2015-10-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106709436B (en) | Track traffic panoramic monitoring-oriented cross-camera suspicious pedestrian target tracking system | |
| CN101950350B (en) | Clear path detection using a hierachical approach | |
| CN105550665B (en) | A kind of pilotless automobile based on binocular vision can lead to method for detecting area | |
| CN102682292B (en) | Method based on monocular vision for detecting and roughly positioning edge of road | |
| CN101656023B (en) | Management method of indoor car park in video monitor mode | |
| CN105005778B (en) | A kind of vehicle on highway detection method of view-based access control model background extracting | |
| Siogkas et al. | Traffic lights detection in adverse conditions using color, symmetry and spatiotemporal information | |
| US5937079A (en) | Method for stereo image object detection | |
| CN101030256B (en) | Method and apparatus for cutting vehicle image | |
| CN107316488B (en) | Signal lamp identification method, device and system | |
| CN107330373A (en) | A kind of parking offense monitoring system based on video | |
| DE102009048892A1 (en) | Clear traveling path detecting method for vehicle e.g. car, involves generating three-dimensional map of features in view based upon preferential set of matched pairs, and determining clear traveling path based upon features | |
| DE102009050492A1 (en) | Travel's clear path detection method for motor vehicle i.e. car, involves monitoring images, each comprising set of pixels, utilizing texture-less processing scheme to analyze images, and determining clear path based on clear surface | |
| DE102009048699A1 (en) | Travel's clear path detection method for motor vehicle i.e. car, involves monitoring images, each comprising set of pixels, utilizing texture-less processing scheme to analyze images, and determining clear path based on clear surface | |
| CN108197523B (en) | Night vehicle detection method and system based on image conversion and contour neighborhood difference | |
| CN109919026B (en) | Surface unmanned ship local path planning method | |
| CN112204614B (en) | Motion segmentation in video from non-stationary cameras | |
| CN109635737A (en) | Automobile navigation localization method is assisted based on pavement marker line visual identity | |
| CN106340031A (en) | Method and device for detecting moving object | |
| CN111967396A (en) | Processing method, device and equipment for obstacle detection and storage medium | |
| CN101369312B (en) | Method and equipment for detecting intersection in image | |
| CN108509900A (en) | Road traffic sign detection and recognition methods, system and storage medium | |
| CN107506753B (en) | Multi-vehicle tracking method for dynamic video monitoring | |
| CN106803073B (en) | Auxiliary driving system and method based on stereoscopic vision target | |
| CN115346155A (en) | Ship image track extraction method for visual feature discontinuous interference |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170118 |
|
| WD01 | Invention patent application deemed withdrawn after publication |