CN110321823A - Zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian - Google Patents
Zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian Download PDFInfo
- Publication number
- CN110321823A CN110321823A CN201910550547.1A CN201910550547A CN110321823A CN 110321823 A CN110321823 A CN 110321823A CN 201910550547 A CN201910550547 A CN 201910550547A CN 110321823 A CN110321823 A CN 110321823A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- illegal
- pavement
- vehicle
- data
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/017—Detecting movement of traffic to be counted or controlled identifying vehicles
- G08G1/0175—Detecting movement of traffic to be counted or controlled identifying vehicles by photographing vehicles, e.g. when violating traffic rules
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Abstract
The invention discloses a kind of, and the zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian, belong to field of image recognition.Zebra stripes based on deep learning of the invention give precedence to illegal its basic step of secondary detection method of pedestrian, read in illegal picture and its lane configurations information, the pedestrian of operating motor vehicles is judged and rejected, identify pavement one skilled in the art and vehicle, again in pavement pedestrian and vehicle subregion count, be based ultimately upon count results and carry out illegal judgement.This method can effectively improve the efficiency of unlawful data screening, and then reach labor-saving effect.By this method to the secondary identification of pedestrian's unlawful data is given precedence to, a large amount of illegal pictures can be filtered out to be artificial.
Description
Technical field
The invention belongs to deep learning fields, and in particular to it is illegal to give precedence to pedestrian to a kind of zebra stripes based on deep learning
Secondary detection method.
Background technique
Zebra stripes are the safety lines of pedestrian's street crossing, and what motor vehicle gave precedence to zebra stripes refraction is the respect to life, are shown
It is urban civilization.National traffic accident 1.4 ten thousand that motor vehicle and pedestrian occurs on zebra stripes altogether, cause 3898 over nearly 3 years
People is dead；In terms of statistical conditions, accident caused by motor vehicle does not give way by regulation accounts for the 90% of national accident.
From the beginning of this year, the whole nation investigates and prosecutes altogether and does not give precedence to zebra stripes illegal activities and rise 3.4 times on year-on-year basis, makes because not giving precedence to zebra stripes
Playing number, death toll at the accident of casualties, respectively thus decline 18% and 9.39% is not difficult to find out on year-on-year basis, zebra stripes order
Concerning life security, treatment action can effectively promote the current safety coefficient of pedestrian and non motorized vehicle, be Xiangcheng City management
Safety engineering that person must pay much attention to, project supported by commen people.Traffic department has also put into effect comity behavior and has newly advised at present, for not giving precedence to
The punishment of 3 points of 50 yuan of fine is detained at behavior.And then traffic accident is effectively reduced.But due to the illegal monitoring camera in current front end
Often there are many wrong bat situations in the technical deficiency of head.So reach the data of traffic hub after illegal monitoring is captured
It generally requires artificial screening and goes out the really illegal data clapped with mistake, this undoubtedly needs huge manpower and material resources.And it manually sieves
Often there are many objective factors in choosing, while also inevitably error.Therefore, it is necessary to propose a kind of comity row based on deep learning
The secondary recognition detection method of people's illegal activities.
Summary of the invention
The purpose of this invention is to solve the problems associated with the prior art, and provides a kind of zebra stripes based on deep learning
Give precedence to the illegal secondary detection method of pedestrian.This method can effectively mitigate the workload of related personnel, and can be improved illegal number
According to accuracy rate prevent artificial fault.
Specific technical solution of the present invention is as follows:
A kind of zebra stripes comity illegal secondary detection method of pedestrian based on deep learning, its step are as follows:
S1: the illegal image data shot by the fixing camera at target crossing is obtained, wherein every group of illegal image data
In enter the first candid photograph picture before the region of pavement containing a vehicle and a vehicle enter behind the region of pavement second
Capture picture；
S2: in the candid photograph ken of the camera pavement region and lane region be marked, wherein lane quantity
For N, and pavement region is divided into N+2 sub-regions, and a pavement is corresponding in the planning driving path in each lane
Subregion, remaining two pavement subregions are located at the two sides in all driving regions；
S3: image data illegal for any group identifies that two are captured row all in picture by target detection model
People, motor vehicle and non-motor vehicle；It is then based on the positional relationship of pedestrian's bounding box and non-motor vehicle bounding box, screen and is rejected is every
The pedestrian for capturing the non-motor vehicle of the driving in picture is opened, pedestrian's data and the vehicle data in every picture are obtained；
S4: the pedestrian's data captured in picture for every obtain pedestrian's foothold in each pedestrian's bounding box, then
Foothold is located at the pedestrian outside the region of pavement to reject, obtains effective pedestrian's data of every candid photograph picture；It is described to have
The pavement subarea number stood in effect pedestrian's data containing each pedestrian；
S5: the vehicle data captured in picture for every calculates each motor vehicle bounding box and pavement region
Overlapping area is accounted for the motor vehicle bounding box area ratio and is judged to driving into people's row more than the motor vehicle of first threshold by overlapping area
Road obtains driving into pavement vehicle data in every candid photograph picture；Described drive into contains each machine in the vehicle data of pavement
The pavement subarea number that Motor Car Institute drives into；
S6: effective pedestrian's data of picture are captured for every in every group of illegal image data respectively and drive into pavement vehicle
Data, in each lane and each pavement subregion pedestrian and vehicle count, obtain:
Pedestrian table data personlist1, record have row contained by each pavement subregion in the first candid photograph picture
People's quantity；
Vehicle table data carlist2, record have the vehicle for driving into pavement in the second candid photograph picture contained by each lane
Quantity；
Pedestrian table data personlist2, record have row contained by each pavement subregion in the second candid photograph picture
People's quantity；
S7: zebra stripes are carried out for every group of illegal image data respectively and give precedence to the illegal identification of pedestrian, if in carlist2
Any one lane LAiIt is middle to there is the vehicle for driving into pavement, then the illegal judgment rule of pedestrian is given precedence to according to zebra stripes, passed through
Whether personlist1 and personlist2 judges vehicle in current lane LAiThe pavement of front or left and right sides
There are still drive into pavement when pedestrian in region；If meeting illegal judgment rule or whether illegal when leaving a question open, which is disobeyed
Method image data is marked and saves, for carrying out subsequent artefacts' judgement.
On the basis of above scheme, the design parameter and way of each step can be used following preferred embodiment and realize.
Preferably, the labeling method in pavement region and lane region is as follows in the step S2:
S21: the lane region in the candid photograph ken of the camera is divided by boundary of lane line, i-th lane note
For LAi, lane sum is N, i=1 ..., N；
S22: for any one lane LAi, the lane line of its two sides is extended, is marked off in the region of pavement
One pavement subregion CR corresponding with the lanei, i=1 ..., N；In addition to the driving region two sides in all lanes, continue
A pavement subregion CR is marked off close to road edge side in the region of pavement0, close to road in the region of pavement
Center line side marks off a pavement subregion CRN+1。
Preferably, being screened from candid photograph picture in the step S3 and rejecting the side for driving the pedestrian of non-motor vehicle
Method are as follows: traverse every two-by-two and capture the pedestrian identified in picture and non-motor vehicle, calculate pedestrian's bounding box and non-motor vehicle side
The overlapping area of boundary's frame, if the ratio that overlapping area accounts for pedestrian's bounding box area is more than high at the top of second threshold and pedestrian's bounding box
When at the top of non-motor vehicle bounding box, determine that the pedestrian for the pedestrian for driving non-motor vehicle, is rejected as dirty data.
Preferably, target detection model uses Yolo model in the step S3.
Preferably, the foothold of each pedestrian selects the midpoint on the bounding box bottom edge of the pedestrian in the step S4.
Preferably, whether pedestrian's foothold is located at outside the region of pavement, by seeking vector cross product in the step S4
Judged.
Preferably, table data personlist1, carlist2, personlist2's obtains in the step S6
Take method as follows:
S61: being read out effective pedestrian's data of the first candid photograph picture, pavement stood based on each pedestrian
Zone number counts pedestrian's quantity in each pavement subregion, constructs pedestrian's table data personlist1=[p0,
p1,…,pN,pN+1], first captures in picture that there are p when pedestrian in k-th of pavement subregionkValue is 1, otherwise value is
0；
S62: the pavement vehicle data that drives into the second candid photograph picture is read out, is sailed based on each motor vehicle
The pavement subarea number entered counts the vehicle fleet size driven into the corresponding pavement subregion in every lane, constructs vehicle
Table data carlist2=[c1,…,cN], wherein the corresponding pavement subregion CR in i-th laneiVehicle is driven into middle presence
When ciValue is 1, otherwise value is 0, i=1 ..., N；
S63: being read out effective pedestrian's data of the second candid photograph picture, pavement stood based on each pedestrian
Zone number counts pedestrian's quantity in each pavement subregion, constructs pedestrian's table data personlist2=[p'0,
p'1,…,p'N,p'N+1], second captures j-th of pavement subregion CR in picturejIn there are p' when pedestrianjValue is 1, on the contrary
Value is 0, j=0,1 ..., N, N+1.
Preferably, carrying out zebra stripes in the step S7 for every group of illegal image data and giving precedence to the illegal knowledge of pedestrian
It is other that the specific method is as follows:
S71: read the corresponding table data personlist1, carlist2 of current illegal image data,
personlist2；
S72: judge in carlist2 with the presence or absence of ci=1, i=1 ..., N, if it exists then successively to each value
For 1 ciS73 is executed, then determines that there is no zebra stripes to give precedence to pedestrian's illegal vehicle in current illegal image data if it does not exist；
S73: the c for being 1 according to valueiSubscript i, from personlist1 extract data [pi-1,pi,pi+1], from
Data [p' is extracted in personlist2i-1,p'i,p'i+1], then carry out following judgement:
If [pi-1,pi,pi+1]=[1,0,0] or [1,1,0], then [p'i-1,p'i,p'i+1As long as] in p'i-1With p'iIn
At least exist a value 1 then determines there are zebra stripes give precedence to pedestrian's illegal vehicle, otherwise be there is no；
If [pi-1,pi,pi+1]=[0,0,1] or [0,1,1], then [p'i-1,p'i,p'i+1As long as] in p'iWith p'i+1In
At least exist a value 1 then determines there are zebra stripes give precedence to pedestrian's illegal vehicle, otherwise be there is no；
If [pi-1,pi,pi+1]=[0,1,0], then [p'i-1,p'i,p'i+1] in work as p'iThen determine that there are spots when value is 1
Horse line give precedence to pedestrian's illegal vehicle, otherwise for there is no；
If [pi-1,pi,pi+1]=[1,0,1] or [1,1,1], then [p'i-1,p'i,p'i+1] in work as p'iWhen value is 1 then
Determine that there are zebra stripes to give precedence to pedestrian's illegal vehicle, works as p'i-1、p'iAnd p'i+1Value is all 0, and there is no zebra stripes to give precedence to row
People's illegal vehicle, other situations are labeled as leaving a question open；
S74: the c for being 1 for each value in carlist2iAfter having executed S73 step, zebra stripes give precedence to row if it exists
People's illegal vehicle is then to the illegal image data of the group labeled as illegal, and zebra stripes are given precedence to pedestrian's illegal vehicle but had if it does not exist
Then to the illegal image data of the group labeled as leaving a question open, remaining situation then is labeled as not disobeying the situation that leaves a question open to the illegal image data of the group
Method.
Preferably, being marked as giving precedence to pedestrian's illegal vehicle there are zebra stripes and being marked as the illegal image to leave a question open
Data are sent to manual examination and verification end, for finally determining that vehicle gives precedence to pedestrian's illegal activities with the presence or absence of zebra stripes.
It is captured by comity pedestrian's violation snap-shooting equipment of front end preferably, the illegal image data calls directly
The illegal image arrived.
The present invention in terms of existing technologies, has the advantages that the present invention can shoot fixing camera
Illegal image data is detected automatically, thus therefrom extract clearly exist or there may be zebra stripes give precedence to pedestrian it is illegal
The image data of vehicle.It is tested using data of this method to artificial screening, the results showed that the illegal data of manual confirmation,
This method can also be determined as unlawful data；The not illegal data of manual confirmation, this method can determine that out that 90% data are
It is not illegal, only 10% mistake.Therefore the efficiency of unlawful data screening can be effectively improved using this method, and then reaches saving
The effect of manpower.By this method to the secondary identification of pedestrian's unlawful data is given precedence to, a large amount of illegal pictures can be filtered out to be artificial.
Detailed description of the invention
Fig. 1 is flow chart of the invention；
Fig. 2 is the first candid photograph picture that vehicle enters before the region of pavement in an embodiment；
Fig. 3 is the second candid photograph picture that vehicle enters behind the region of pavement in an embodiment；
Fig. 4 is that the lane at 2 lane crossings and pavement divide schematic diagram in the present embodiment；
Fig. 5 is pedestrian's dirty data of the driving motorcycles identified；
Fig. 6 is the pedestrian area that detected；
Fig. 7 is the pedestrian that detected and motor vehicles；
Fig. 8 is the pedestrian's bounding box and motor vehicles bounding box into pavement region that detected；
Fig. 9 is that the lane at 1 lane crossing and pavement divide schematic diagram in another embodiment；
Figure 10 is the first candid photograph picture that 1 lane crossing vehicle enters before the region of pavement in another embodiment；
Figure 11 is the second candid photograph picture that 1 lane crossing vehicle enters behind the region of pavement in another embodiment.
Specific embodiment
The present invention is further elaborated and is illustrated with reference to the accompanying drawings and detailed description.Each implementation in the present invention
The technical characteristic of mode can carry out the corresponding combination under the premise of not conflicting with each other.
As shown in Figure 1, the zebra stripes of the invention based on deep learning give precedence to the illegal secondary detection method of pedestrian, it is basic
Step is to read in illegal picture and its lane configurations information, and the pedestrian of operating motor vehicles is judged and rejected, and identifies people's row
Road one skilled in the art and vehicle, then in pavement pedestrian and vehicle subregion count, be based ultimately upon count results carry out it is illegal
Determine.The process of its realization is illustrated below by embodiment.
In the present embodiment, zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian, are mainly used for pair
The illegal image data of fixing camera shooting carries out preliminary screening, excludes the image data that illegal activities are obviously not present, with
Reduce the workload of manual examination and verification identification.At present violation snap-shootings equipment, the parts such as electronic eyes are largely installed in city to set
It is standby to have the function of to give precedence to pedestrian's violation snap-shooting, it can tentatively capture to obtain the illegal image evidence that vehicle does not give precedence to pedestrian.But by
In the technical deficiency of the illegal monitoring camera in current front end, often exist it is many it is wrong clap situations, therefore method of the invention can be with
The illegal image captured by comity pedestrian's violation snap-shooting equipment of front end is called directly, secondary detection is carried out to it, is screened out
Dirty data.
Zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian, and specific step is as follows:
S1: the illegal image data shot by the camera at target crossing is obtained, wherein containing in every group of illegal image data
The first candid photograph picture and a vehicle before having a vehicle to enter pavement region enter the second candid photograph behind the region of pavement
Picture.
For camera in the present embodiment using fixed fixed point camera, installation site, shooting angle, acquisition parameters are equal
It keeps identical, thereby guarantees that the background (including lane line position) in shooting image other than vehicle also keeps identical.Such as Fig. 2
With shown in Fig. 3, for wherein one group of unlawful data (the as front end violation snap-shooting camera for importing same bayonet number in the present embodiment
The unlawful data of candid photograph), it is made of two illegal pictures, Fig. 2 is that vehicle enters the first candid photograph picture before the region of pavement, figure
3 enter the second candid photograph picture behind the region of pavement for vehicle.
S2: to the pavement area in the candid photograph ken (i.e. by the sensor captured image range of camera) of the camera
Domain and lane region are marked, and since camera position will not change for a long time, therefore only need to mark for the camera of unified bayonet
Note is primary, it will be able to secondary detection identification is carried out to the unlawful data captured under this camera, if camera is because of maintenance etc.
There is movement and then needs to re-start label in reason.If wherein the lane quantity at the crossing is N, then pavement region needs
N+2 sub-regions are divided into, a pavement subregion, remaining two people's rows are corresponding in the planning driving path in each lane
Road subregion is located at the two sides in all driving regions.The basic process of the labeling method in pavement region and lane region is such as
Under:
S21: the lane region in the candid photograph ken of the camera is divided by boundary of lane line, i-th lane note
For LAi, lane sum is N, i=1 ..., N；
S22: for any one lane LAi, the lane line of its two sides is extended, is marked off in the region of pavement
One pavement subregion CR corresponding with the lanei, i=1 ..., N.In addition to the driving region two sides in all lanes, continue
A pavement subregion CR is marked off close to road edge (curb position) side in the region of pavement0, in pavement region
In close to road axis (double amber lines or single yellow line) side mark off a pavement subregion CRN+1.Wherein CR0It represents
Pedestrian not yet starts to pass through zebra stripes or just region of the station at zebra stripes initial position, CRN+It represents pedestrian and has passed over one
All lanes in direction are prepared to enter into opposite lane and correspond to region locating when pavement, probably in the lane of both direction
The two sides in heart line of demarcation, motor vehicle should also be as being given precedence to when in the two regions.
It in the present embodiment,, can be first by pavement region with not in real marking by taking the 2 lane crossings of Fig. 4 as an example
Regular quadrilateral marks (vertex is four points: x1, x5, x6, x10), the coboundary in pavement region and lower boundary in image
Respectively two stop lines at crossing, left border are the left side edge curb of road, and right side boundary is center double amber lines.Then
People's row region is divided according to the extended line in lane again, while the pavement subregion CR by reserved lane0And CRN+1.Figure
Share 2 lanes in 4, therefore pavement region division is four regions, therefore need six points: x2, x3, x4, x7, x8, x9 are carried out
It divides.Pavement region and lane region such as Fig. 4 after final label.The wherein corresponding CR in region 10, region 2 and 3 point half corresponding 2
Lane, the corresponding CR in region 4N+1。
So far we have obtained two illegal image datas and lane, pavement configuration information (mark shown in Fig. 4
10 coordinate informations arrived), it will be able to identification decision is carried out to comity pedestrian's unlawful data by the method for the invention.
S3: image data illegal for any group identifies that two are captured row all in picture by target detection model
People, motor vehicle and non-motor vehicle.The target detection model being loaded into the present embodiment is Yolo model, is adjusted to Yolo model
So as to identify 3 kinds of target categories that we need, it may be assumed that pedestrian, non-motor vehicle, motor vehicle and their position letter
It ceases (bounding box, bounding box).
It is then based on the positional relationship of pedestrian's bounding box and non-motor vehicle bounding box, screen and is rejected in every candid photograph picture
Driving non-motor vehicle pedestrian, obtain pedestrian's data and the vehicle data in every picture；
By back we can obtain pedestrian all in picture, motor vehicle, non-motor vehicle location information.According to
The illegal regulation of pedestrian is given precedence to, motor vehicle does not need to give precedence to the pedestrian for riding non-motor vehicle.So being identified using target detection model
Pedestrian out can have the pedestrian that rides, and we term it pedestrian's dirty data (the as shown in the figure 5 pedestrian's data that ride motorcycle).
Although target detection model is the pedestrian that cannot recognize that operating motor vehicles, motor vehicle and pedestrian can be identified respectively,
So can be determined based on the positional relationship of pedestrian's bounding box and non-motor vehicle bounding box, meet before regional scope is then determined as
The dirty data that face is mentioned, pedestrian's data invalid do not make final illegal judgement.The removal of dirty data in the present embodiment, i.e., from grabbing
Clap the method that the pedestrian for driving non-motor vehicle is screened and rejected in picture are as follows: traverse every two-by-two and capture the row identified in picture
People and non-motor vehicle calculate the overlapping area of pedestrian's bounding box and non-motor vehicle bounding box, if overlapping area accounts for pedestrian's bounding box
The ratio of area is more than to determine the pedestrian when being higher than at the top of non-motor vehicle bounding box at the top of threshold value A and pedestrian's bounding box to drive
The pedestrian of non-motor vehicle, is rejected as dirty data.The specific value of threshold value A can be excellent according to being actually adjusted
Change, herein preferably 50%, thus rejects pedestrian's data of driving motorcycles in Fig. 5.
Pedestrian and its location information are had been detected by by this step, and filters out the pedestrian that rides, whole has been obtained and has disobeyed
Real pedestrian's data, each pedestrian are denoted as person (left in method picture1, top1, right1, bottom1), (left1,
top1) be pedestrian's bounding box top left corner apex coordinate, (right1, bottom1) be pedestrian's bounding box bottom right angular vertex
Coordinate.But since pedestrian's data are the data in whole picture, part pedestrian, therefore need to be into one not in the region of pavement
Step screens pedestrian's data, finds out the pedestrian in pavement region (x1, x5, x6,4 points of regions surrounded x10 in Fig. 4)
Data.
Likewise, also having obtained vehicle position information in whole illegal picture in this step, each vehicle is denoted as car
(left2, top2, right2, bottom2), (left2, top2) be the vehicle bounding box top left corner apex coordinate, (right2,
bottom2) be the vehicle bounding box lower right corner apex coordinate.
S4: since the pedestrian area (such as Fig. 6 and 7) that detected is irregular, if entering pavement using pedestrian
Area, which accounts for, detects that pedestrian demarcates the ratio of frame area as judgment basis, it is clear that does not meet convention.For such situation, how
The pedestrian is determined whether in pavement, and this method uses and only takes the pedestrian's coordinate data detected in target detection model
(left1, top1, right1, bottom1) in c point (foothold for being defined as pedestrian), by c point replace pedestrian position judgement be
It is no in the region of pavement, if not rejected then being then stored in.The foothold of each pedestrian selects the bounding box bottom of the pedestrian
The midpoint on side, i.e. (left1+right1/ 2, bottom) point.Specific practice are as follows:
The pedestrian's data captured in picture for every, obtain pedestrian's foothold in each pedestrian's bounding box, then will
The pedestrian that foothold is located at outside the region of pavement rejects, and obtains effective pedestrian's data of every candid photograph picture.Pedestrian stops over
Point whether be located at pavement region outside, can be judged in several ways, one way in which be by ask vector cross product into
Row judgement, resolution principle are the point inside convex polygon all in the same side of the vector where the side of convex polygon, are pitched using asking
Product:
Assuming that four vertex of quadrangle are followed successively by A (x1, y1), B (x2, y2), C (x3, y3), D (x4, y4) need to judge
Point be P (x, y), if fruit dot P is inside quadrangle, then vector AB*AP (here it is cross product is sought, is also equal to (x2-x1) * (y-
Y1)-(y2-y1) * (x-x1)) value and BC*BP, CD*CP, DA*DP value jack per line (if there is null situation, then it represents that P exists
On side).I.e. four values are positive together or with being negative, then point P is inside ABCD, otherwise in outside.
By this step, the pedestrian for not standing on pavement region can be rejected, retains and finally really stands on people's row
The pedestrian in road region, these pedestrians are the pedestrians for preparing to go across the road or going across the road, and motor vehicle needs to give precedence to it.
In order to which subsequent statistical is convenient, need to record pavement that each pedestrian is stood in this step in the effective pedestrian's data saved
The data such as zone number and the bounding box of pedestrian.
S5: the vehicle data as obtained in S3 is the data in whole picture, need to screen, find out to vehicle data
The vehicle data of (x1, x5, x6, x10 surround region in Fig. 4) is driven into the region of pavement.Vehicle position information can be substituted into
In the region of pavement, judges whether to enter pavement, the vehicle data is then retained into pavement, otherwise removes the vehicle number
According to.Unlike but, the method that vehicle cannot enter in pavement according to pedestrian be determined.Due to the positioning of vehicle and people
There are larger differences, so judging whether vehicle enters pavement with " foothold " of pedestrian's data is illogicality
's.Therefore pavement area is driven into using vehicle herein and judges whether vehicle enters pavement region, specific practice are as follows:
The vehicle data captured in picture for every calculates being overlapped for each motor vehicle bounding box and pavement region
Overlapping area is accounted for the motor vehicle that the motor vehicle bounding box area ratio is more than threshold value B and is judged to driving into pavement, obtained by area
It captures in picture for every and drives into pavement vehicle data.The specific value of threshold value B can be according to being actually adjusted optimization, this
Place preferably 30% regards as vehicle entrance when the bounding box area that vehicle accounts for vehicle with pavement overlapping area is more than 30%
Pavement.According to this decision rule, the vehicle data for driving into pavement is retained, does not drive into, removes.It thus will be above Fig. 8
Three vehicles rejected.
In order to which subsequent statistical is convenient, the pavement sub-district driven into the vehicle data of pavement containing each motor vehicle is driven into
The data such as the bounding box of Field Number, the corresponding lane number of pavement subregion and motor vehicle.
S6: effective pedestrian's data of picture are captured for every in every group of illegal image data respectively and drive into pavement vehicle
Data, in each lane and each pavement subregion pedestrian and vehicle count, obtain three kinds of table datas,
It is respectively as follows:
Pedestrian table data personlist1, record has each pavement subregion in the first candid photograph picture in the list
Contained pedestrian's quantity；
Vehicle table data carlist2, record, which has in the second candid photograph picture, in the list drives into people contained by each lane
The vehicle fleet size on trade；
Pedestrian table data personlist2, record has each pavement subregion in the second candid photograph picture in the list
Contained pedestrian's quantity；
The acquisition methods of above-mentioned table data personlist1, carlist2, personlist2 are as follows:
S61: being read out effective pedestrian's data of the first candid photograph picture, pavement stood based on each pedestrian
Zone number counts pedestrian's quantity in each pavement subregion, constructs pedestrian's table data personlist1=[p0,
p1,…,pN,pN+1], first captures in picture that there are p when pedestrian in k-th of pavement subregionkValue is 1, otherwise value is
0；
S62: the pavement vehicle data that drives into the second candid photograph picture is read out, is sailed based on each motor vehicle
The pavement subarea number entered counts the vehicle fleet size driven into the corresponding pavement subregion in every lane, constructs vehicle
Table data carlist2=[c1,…,cN], wherein the corresponding pavement subregion CR in i-th laneiVehicle is driven into middle presence
When ciValue is 1, otherwise value is 0, i=1 ..., N；
S63: being read out effective pedestrian's data of the second candid photograph picture, pavement stood based on each pedestrian
Zone number counts pedestrian's quantity in each pavement subregion, constructs pedestrian's table data personlist2=[p'0,
p'1,…,p'N,p'N+1], second captures j-th of pavement subregion CR in picturejIn there are p' when pedestrianjValue is 1, on the contrary
Value is 0, j=0,1 ..., N, N+1.
By taking the crossing of Fig. 4 as an example, in the present embodiment, pavement region division is gone out into region 1, region 2, region 3 and area
Domain 4.Therefore region 2 and region 3 are divided into vehicle data according to the rule that pedestrian, vehicle enter pavement, are counted respectively
Number, that is, pavement vehicle data will be driven into and be divided into two groups, with carlist=[c1,c2] indicate, c1,c2Value is 0 or 1.
If there are vehicle then c in region 21Value is 1, if vehicle c is not present in region 21Value is 0.Likewise, if there are vehicles in region 3
Then c2Value is 1, if vehicle c is not present in region 32Value is 0.Similarly effective pedestrian's data are divided into region 1, region 2, region 3
With region 4, first candid photograph picture personlist1=[p0,p1,p2,p3] indicate, second candid photograph picture is used
Personlist2=[p'0,p'1,p'2,p'3] indicate, p0,p1,p2,p3,p'0,p'1,p'2,p'3Value is that 0 or 1,1 expression should
Pavement subregion there are pedestrian, 0 for there is no.After aforementioned processing, (two illegal for image data illegal for one group
Picture) obtain following data: (carlist1, the personlist1) of the first picture, the second picture (carlist2,
personlist2).Wherein carlist1 and carlist2 be length be 2 one-dimensional list, personlist1 and
Personlist2 is the one-dimensional list that length is 4.Certainly, in subsequent decision process, carlist1 is not used actually.
S7: zebra stripes are carried out for every group of illegal image data respectively and give precedence to the illegal identification of pedestrian, if in carlist2
Any one lane LAiIt is middle to there is the vehicle for driving into pavement, then the illegal judgment rule of pedestrian is given precedence to according to zebra stripes, passed through
Whether personlist1 and personlist2 judges vehicle in current lane LAiThe pavement of front or left and right sides
There are still drive into pavement when pedestrian in region；If meeting illegal judgment rule or whether illegal when leaving a question open, which is disobeyed
Method image data is marked and saves, for carrying out subsequent artefacts' judgement.
The specific method that zebra stripes give precedence to the illegal identification of pedestrian is carried out for every group of illegal image data, it can be by as follows
Sub-step is realized:
S71: read the corresponding table data personlist1, carlist2 of current illegal image data,
personlist2；
S72: judge in carlist2 with the presence or absence of ci=1, i=1 ..., N, if it exists then successively to each value
For 1 ciS73 is executed, then determines that there is no zebra stripes to give precedence to pedestrian's illegal vehicle in current illegal image data if it does not exist；
S73: the c for being 1 according to valueiSubscript i, from personlist1 extract data [pi-1,pi,pi+1], from
Data [p' is extracted in personlist2i-1,p'i,p'i+1], then carry out following judgement:
If [pi-1,pi,pi+1]=[1,0,0] or [1,1,0], then [p'i-1,p'i,p'i+1As long as] in p'i-1With p'iIn
At least exist a value 1 then determines there are zebra stripes give precedence to pedestrian's illegal vehicle, otherwise be there is no；
If [pi-1,pi,pi+1]=[0,0,1] or [0,1,1], then [p'i-1,p'i,p'i+1As long as] in p'iWith p'i+1In
At least exist a value 1 then determines there are zebra stripes give precedence to pedestrian's illegal vehicle, otherwise be there is no；
If [pi-1,pi,pi+1]=[0,1,0], then [p'i-1,p'i,p'i+1] in work as p'iThen determine that there are spots when value is 1
Horse line give precedence to pedestrian's illegal vehicle, otherwise for there is no；
If [pi-1,pi,pi+1]=[1,0,1] or [1,1,1], then [p'i-1,p'i,p'i+1] in work as p'iWhen value is 1 then
Determine that there are zebra stripes to give precedence to pedestrian's illegal vehicle, works as p'i-1、p'iAnd p'i+1Value is all 0, and there is no zebra stripes to give precedence to row
People's illegal vehicle, other situations are labeled as leaving a question open；
S74: the c for being 1 for each value in carlist2iAfter having executed S73 step, zebra stripes give precedence to row if it exists
People's illegal vehicle is then to the illegal image data of the group labeled as illegal, and zebra stripes are given precedence to pedestrian's illegal vehicle but had if it does not exist
Then to the illegal image data of the group labeled as leaving a question open, remaining situation then is labeled as not disobeying the situation that leaves a question open to the illegal image data of the group
Method.
Since when the lane quantity difference at crossing, decision process is had differences.For it easier comprehension, we are scheming
On the basis of 4 two-way traffic crossing, it is re-introduced into another the image of crossing for only having a lane, as shown in Figure 9.For the crossing
Illegal image data, the personlist1 of our available first pictures, the carlist2 of the second picture,
personlist2.Wherein carlist2=[c1] it is the one-dimensional list (c that length is 11Show pavement in front of lane for 1
Indicate to be not present there are vehicle in region, 0), personlist1=[p0,p1,p2] and personlist2=[p'0,p'1,p'2]
It is the one-dimensional list that length is 3.
A. we discuss the unlawful data for a lane (there was only 1 lane, Fig. 9 in unlawful data) first:
The illegal decision rule of pedestrian is given precedence to according to zebra stripes: being captured in first illegal picture when motor vehicle enters camera
Region, if the motor vehicle is located in the lane or to adjacent lane memory in lane in pedestrian.So in second illegal figure
Candid photograph enters pavement to illegal vehicle in piece and pedestrian does not pass through lane where the illegal vehicle, and pedestrian is still original
In lane, then the vehicle illegal.One group of unlawful data as shown in Figure 10 and Figure 11 should be determined to have separated according to the rule
Method vehicle.
It is as follows that analytical procedure is carried out using the method for step S7 of the invention:
1. when there are illegal vehicle, c in pavement region in the illegal picture of carlist2 i.e. second1Value is 1, then right
Personlist1 and personlist2 carries out analysis and enters step 2, and on the contrary then analysis is there is no illegal vehicles.
2. according to personlist1=[p0,p1,p2] and personlist2=[p'0,p'1,p'2] in each element value
Determined:
When data are [1,0,0] or [1,1,0] in pesonlist1, then personlist2=[p'0,p'1,p'2]
As long as middle data p'0With p'1Middle value at least has one 1 and determines to give precedence to pedestrian's illegal vehicle there are zebra stripes, otherwise for not
There are zebra stripes to give precedence to pedestrian's illegal vehicle.
When data are [0,0,1] or [0,1,1] in pesonlist1, then personlist2=[p'0,p'1,p'2]
As long as middle data p'1With p'2Middle value at least has one 1 and determines to give precedence to pedestrian's illegal vehicle there are zebra stripes, otherwise for not
There are zebra stripes to give precedence to pedestrian's illegal vehicle.
When data are [0,1,0] in pesonlist1, then personlist2=[p'0,p'1,p'2] in data only work as
p'1Value is still 1 and determines to give precedence to pedestrian's illegal vehicle there are zebra stripes, otherwise is that there is no zebra stripes to give precedence to the illegal vehicle of pedestrian
?.
When data are [1,0,1] in pesonlist1, then personlist2=[p'0,p'1,p'2] in data only work as
p'1Value is still 1 and determines that there are zebra stripes to give precedence to pedestrian's illegal vehicle, works as p'0,p'1,p'2Middle value is all 0 and is not present
Zebra stripes give precedence to pedestrian's illegal vehicle, other situations are to leave a question open.
When data are [1,1,1] in pesonlist1, then personlist2=[p'0,p'1,p'2] in data only work as
p'1Value is still 1 and determines that there are zebra stripes to give precedence to pedestrian's illegal vehicle, works as p'0,p'1,p'2Middle value is all 0 and is not present
Zebra stripes give precedence to pedestrian's illegal vehicle, other situations are to leave a question open.
3. passing through step 1, step 2, we can be obtained in group candid photograph data with the presence or absence of illegal vehicle, for depositing
Doubtful vehicle can be by manually carrying out further verifying judgement.
B. according to a lane determination method principle, available multilane determination method.It is with the two lane highways in Fig. 4 below
Example is illustrated:
According to aforementioned step, we have obtained the personlist1=[p of the first picture0,p1,p2,p3], second
Carlist2=[the c of picture1,c2], personlist2=[p'0,p'1,p'2,p'3].It is 2 that wherein carlist2, which is length,
One-dimensional list, personlist1 and personlist2 are the one-dimensional lists that length is 4.
It is as follows that analytical procedure is carried out using the method for step S7 of the invention:
1. then being carried out to personlist1 and personlist2 when carlist2 is [1,0] or [0,1] or [1,1]
Analysis enters step 2, and on the contrary then analysis is there is no illegal vehicles.
2. according to carlist2=[c1,c2] in each element value determined:
When carlist2 is [1,0] or [0,1], then by personlist1=[p0,p1,p2,p3], personlist2=
[p'0,p'1,p'2,p'3] it is split as 1 lane data: if carlist2 takes personlist1 and personlist2 when being [1,0]
Preceding 3 data personlist1=[p0,p1,p2], personlist2=[p'0,p'1,p'2] substitute into aforementioned 1 lane judgement
Judge in method.With should carlist2 be [0,1], that is, take rear 3 data of personlist1 and personlist2
Personlist1=[p1,p2,p3], personlist2=[p'1,p'2,p'3] substitute into, it is judged whether there is with this illegal.
3. when carlist2 is [1,1], equally by personlist1=[p0,p1,p2,p3], personlist2=[p'0,
p'1,p'2,p'3] it is split as 1 lane data:
Preceding 3 data of personlist1 and personlist2 is first taken to substitute into 1 lane determination method, judgement is
No illegal judgement result zq1 (three kinds of value point: 0- is illegal, and 1- is not illegal, and 2- leaves a question open).
Then it takes rear 3 data of personlist1 and personlist2 to substitute into 1 lane determination method again, judges
To whether illegal judgement result zq2 (three kinds of value point: 0- is illegal, and 1- is not illegal, and 2- leaves a question open).
When there are a values 1 then to be illegal in zq1 or zq2；When, there are a value 2, there is no take in zq1 or zq2
Value 1 then leaves a question open；Remaining situation is not illegal.
Certainly, if crossing has lanes more more than two lanes, judged also based on identical principle.
It is marked as giving precedence to pedestrian's illegal vehicle there are zebra stripes and is marked as the illegal image data to leave a question open, be both needed to
It is sent to manual examination and verification end, for finally determining that vehicle gives precedence to pedestrian's illegal activities with the presence or absence of zebra stripes.
In order to test detection accuracy of the invention, the above method of the invention is applied to have already passed through artificial judgment
In data set.The result shows that the image data illegal for manual confirmation, this method can be also determined as illegal number
According to；For the not illegal image data of manual confirmation, this method will determine 90% image data be it is not illegal, only
10% mistake.Therefore, after being screened using this method, subsequent artificial judgement only need to judgement result of the invention into
Row further confirms that, be discharged from decision error as a result, this will effectively improve the efficiency of unlawful data screening, and then reach section
The effect of human-saving.
Above-mentioned embodiment is only a preferred solution of the present invention, so it is not intended to limiting the invention.Have
The those of ordinary skill for closing technical field can also make various changes without departing from the spirit and scope of the present invention
Change and modification.Therefore all mode technical solutions obtained for taking equivalent substitution or equivalent transformation, all fall within guarantor of the invention
It protects in range.
Claims (10)
1. a kind of zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian, which is characterized in that steps are as follows:
S1: the illegal image data shot by the fixing camera at target crossing is obtained, wherein containing in every group of illegal image data
The first candid photograph picture and a vehicle before having a vehicle to enter pavement region enter the second candid photograph behind the region of pavement
Picture；
S2: in the candid photograph ken of the camera pavement region and lane region be marked, wherein lane quantity be N,
And pavement region is divided into N+2 sub-regions, and a pavement sub-district is corresponding in the planning driving path in each lane
Domain, remaining two pavement subregions are located at the two sides in all driving regions；
S3: image data illegal for any group identifies that two are captured pedestrian all in picture, machine by target detection model
Motor-car and non-motor vehicle；It is then based on the positional relationship of pedestrian's bounding box and non-motor vehicle bounding box, screens and rejects every and grab
The pedestrian for clapping the driving non-motor vehicle in picture, obtains pedestrian's data and the vehicle data in every picture；
S4: the pedestrian's data captured in picture for every obtain pedestrian's foothold in each pedestrian's bounding box, then will fall
The pedestrian that pin point is located at outside the region of pavement rejects, and obtains effective pedestrian's data of every candid photograph picture；Effective row
The pavement subarea number stood in personal data containing each pedestrian；
S5: the vehicle data captured in picture for every calculates being overlapped for each motor vehicle bounding box and pavement region
Overlapping area is accounted for the motor vehicle bounding box area ratio and is judged to driving into pavement more than the motor vehicle of first threshold by area,
It obtains driving into pavement vehicle data in every candid photograph picture；Described drive into contains each motor vehicle in the vehicle data of pavement
The pavement subarea number driven into；
S6: effective pedestrian's data of picture are captured for every in every group of illegal image data respectively and drive into pavement vehicle number
According to, in each lane and each pavement subregion pedestrian and vehicle count, obtain:
Pedestrian table data personlist1, record have pedestrian's number contained by each pavement subregion in the first candid photograph picture
Amount；
Vehicle table data carlist2, record have the vehicle number for driving into pavement in the second candid photograph picture contained by each lane
Amount；
Pedestrian table data personlist2, record have pedestrian's number contained by each pavement subregion in the second candid photograph picture
Amount；
S7: carrying out zebra stripes for every group of illegal image data respectively and give precedence to the illegal identification of pedestrian, if appointing in carlist2
One lane LAiIt is middle to there is the vehicle for driving into pavement, then the illegal judgment rule of pedestrian is given precedence to according to zebra stripes, passed through
Whether personlist1 and personlist2 judges vehicle in current lane LAiThe pavement of front or left and right sides
There are still drive into pavement when pedestrian in region；If meeting illegal judgment rule or whether illegal when leaving a question open, which is disobeyed
Method image data is marked and saves, for carrying out subsequent artefacts' judgement.
2. the zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian as described in claim 1, feature exists
In in the step S2, the labeling method in pavement region and lane region is as follows:
S21: the lane region in the candid photograph ken of the camera is divided by boundary of lane line, and i-th lane is denoted as LAi,
Lane sum is N, i=1 ..., N；
S22: for any one lane LAi, the lane line of its two sides is extended, one is marked off in the region of pavement
Pavement subregion CR corresponding with the lanei, i=1 ..., N；In addition to the driving region two sides in all lanes, continue in people's row
A pavement subregion CR is marked off close to road edge side in road region0, close to road-center in the region of pavement
Line side marks off a pavement subregion CRN+1。
3. the zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian as described in claim 1, feature exists
In being screened in picture from capturing and reject the method for driving the pedestrian of non-motor vehicle are as follows: traversal is every two-by-two in the step S3
It opens and captures the pedestrian identified in picture and non-motor vehicle, calculate the overlapping area of pedestrian's bounding box and non-motor vehicle bounding box,
If the ratio that overlapping area accounts for pedestrian's bounding box area is more than to be higher than non-motor vehicle boundary at the top of second threshold and pedestrian's bounding box
When arch, determine that the pedestrian for the pedestrian for driving non-motor vehicle, is rejected as dirty data.
4. the zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian as described in claim 1, feature exists
In in the step S3, target detection model uses Yolo model.
5. the zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian as described in claim 1, feature exists
In in the step S4, the foothold of each pedestrian selects the midpoint on the bounding box bottom edge of the pedestrian.
6. the zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian as described in claim 1, feature exists
In in the step S4, whether pedestrian's foothold is located at outside the region of pavement, by asking vector cross product to be judged.
7. the zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian as described in claim 1, feature exists
In in the step S6, the acquisition methods of table data personlist1, carlist2, personlist2 are as follows:
S61: effective pedestrian's data of the first candid photograph picture are read out, the pavement subregion stood based on each pedestrian
Number counts pedestrian's quantity in each pavement subregion, constructs pedestrian's table data personlist1=[p0,p1,…,
pN,pN+1], first captures in picture that there are p when pedestrian in k-th of pavement subregionkValue is 1, otherwise value is 0；
S62: the pavement vehicle data that drives into the second candid photograph picture is read out, is driven into based on each motor vehicle
Pavement subarea number counts the vehicle fleet size driven into the corresponding pavement subregion in every lane, constructs vehicle stock
Table data carlist2=[c1,…,cN], wherein the corresponding pavement subregion CR in i-th laneiWhen vehicle is driven into middle presence
ciValue is 1, otherwise value is 0, i=1 ..., N；
S63: effective pedestrian's data of the second candid photograph picture are read out, the pavement subregion stood based on each pedestrian
Number counts pedestrian's quantity in each pavement subregion, constructs pedestrian's table data personlist2=[p'0,p
'1,…,p'N,p'N+1], second captures j-th of pavement subregion CR in picturejIn there are p' when pedestrianjValue is 1, on the contrary
Value is 0, j=0,1 ..., N, N+1.
8. the zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian as described in claim 1, feature exists
In carrying out zebra stripes for every group of illegal image data and give precedence to the specific method of the illegal identification of pedestrian such as in the step S7
Under:
S71: current illegal image data corresponding table data personlist1, carlist2, personlist2 are read；
S72: judge in carlist2 with the presence or absence of ci=1, i=1 ..., N are successively then if it exists 1 to each value
ciS73 is executed, then determines that there is no zebra stripes to give precedence to pedestrian's illegal vehicle in current illegal image data if it does not exist；
S73: the c for being 1 according to valueiSubscript i, from personlist1 extract data [pi-1,pi,pi+1], from
Data [p' is extracted in personlist2i-1,p'i,p'i+1], then carry out following judgement:
If [pi-1,pi,pi+1]=[1,0,0] or [1,1,0], then [p'i-1,p'i,p'i+1As long as] in p'i-1With p'iIn at least
There are a value 1 then determine there are zebra stripes give precedence to pedestrian's illegal vehicle, otherwise for there is no；
If [pi-1,pi,pi+1]=[0,0,1] or [0,1,1], then [p'i-1,p'i,p'i+1As long as] in p'iWith p'i+1In at least
There are a value 1 then determine there are zebra stripes give precedence to pedestrian's illegal vehicle, otherwise for there is no；
If [pi-1,pi,pi+1]=[0,1,0], then [p'i-1,p'i,p'i+1] in work as p'iThen determine that there are zebra stripes when value is 1
Give precedence to pedestrian's illegal vehicle, otherwise for there is no；
If [pi-1,pi,pi+1]=[1,0,1] or [1,1,1], then [p'i-1,p'i,p'i+1] in work as p'iValue then determines when being 1
There are zebra stripes to give precedence to pedestrian's illegal vehicle, works as p'i-1、p'iAnd p'i+1Value is all 0, and there is no zebra stripes comity pedestrians to disobey
Method vehicle, other situations are labeled as leaving a question open；
S74: the c for being 1 for each value in carlist2iAfter having executed S73 step, zebra stripes are given precedence to pedestrian and are disobeyed if it exists
Method vehicle is then to the illegal image data of the group labeled as illegal, and zebra stripes are given precedence to pedestrian's illegal vehicle but had and leave a question open if it does not exist
Situation is then to the illegal image data of the group labeled as leaving a question open, remaining situation is not then to the illegal image data of the group labeled as illegal.
9. the zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian as described in claim 1, feature exists
In being marked as giving precedence to there are zebra stripes and pedestrian's illegal vehicle and be marked as the illegal image data to leave a question open, be sent to
Manual examination and verification end, for finally determining that vehicle gives precedence to pedestrian's illegal activities with the presence or absence of zebra stripes.
10. the zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian as described in claim 1, feature exists
In the illegal image data calls directly the illegal image captured by comity pedestrian's violation snap-shooting equipment of front end.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910550547.1A CN110321823B (en) | 2019-06-24 | 2019-06-24 | Zebra crossing gift pedestrian illegal secondary detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910550547.1A CN110321823B (en) | 2019-06-24 | 2019-06-24 | Zebra crossing gift pedestrian illegal secondary detection method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110321823A true CN110321823A (en) | 2019-10-11 |
| CN110321823B CN110321823B (en) | 2021-03-02 |
Family
ID=68121130
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910550547.1A Active CN110321823B (en) | 2019-06-24 | 2019-06-24 | Zebra crossing gift pedestrian illegal secondary detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110321823B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110675637A (en) * | 2019-10-15 | 2020-01-10 | 上海眼控科技股份有限公司 | Vehicle illegal video processing method and device, computer equipment and storage medium |
| CN111508082A (en) * | 2019-10-31 | 2020-08-07 | 杭州海康威视系统技术有限公司 | Parking information acquisition method and device, central platform and storage medium |
| CN111754773A (en) * | 2020-07-01 | 2020-10-09 | 宁波工程学院 | Road violation monitoring configuration method and system, camera and storage medium |
| CN112446334A (en) * | 2020-12-02 | 2021-03-05 | 福建亿安智能技术有限公司 | Method and system for recognizing illegal behaviors of non-motor vehicle |
| CN112949559A (en) * | 2020-10-23 | 2021-06-11 | 深圳巴士集团股份有限公司 | Pedestrian gift detection method and device and terminal equipment |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20090010259A (en) * | 2007-07-20 | 2009-01-30 | 전병태 | System and method for regulating illegal headlight vehicles |
| CN102592450A (en) * | 2012-02-22 | 2012-07-18 | 杭州海康威视数字技术股份有限公司 | System and method for snapshotting behavior of motor vehicles not giving precedence to pedestrians on zebra stripes |
| CN104361747A (en) * | 2014-11-11 | 2015-02-18 | 杭州新迪数字工程系统有限公司 | Automatic capture system and recognition method for vehicles not giving way to passengers on zebra crossing |
| CN107730906A (en) * | 2017-07-11 | 2018-02-23 | 银江股份有限公司 | Zebra stripes vehicle does not give precedence to the vision detection system of pedestrian behavior |
| CN108090457A (en) * | 2017-12-27 | 2018-05-29 | 天津天地人和企业管理咨询有限公司 | A kind of motor vehicle based on video does not give precedence to pedestrian detection method |
| CN108447305A (en) * | 2018-04-21 | 2018-08-24 | 上海交通大学 | A kind of new friendship rule are lower to give precedence to pedestrian's method for early warning and early warning system |
| CN108847032A (en) * | 2018-08-21 | 2018-11-20 | 北京深瞐科技有限公司 | A kind of traffic violation recognition methods and device |
| CN109035800A (en) * | 2018-09-30 | 2018-12-18 | 天津天地人和企业管理咨询有限公司 | A kind of method that intelligent transportation video camera automatic detecting machine motor-car does not avoid pedestrian |
| CN109300313A (en) * | 2017-07-25 | 2019-02-01 | 杭州海康威视数字技术股份有限公司 | Detection method, camera and the server of illegal activities |
| CN109409191A (en) * | 2018-08-24 | 2019-03-01 | 广东智媒云图科技股份有限公司 | A kind of zebra stripes vehicle evacuation detection method and system based on machine learning |
| CN109493609A (en) * | 2018-12-11 | 2019-03-19 | 杭州炬视科技有限公司 | A kind of portable device and method for not giving precedence to the candid photograph of pedestrian's automatic identification |
| CN109766743A (en) * | 2018-11-21 | 2019-05-17 | 汪明智 | A kind of intelligent bionic policing system |
-
2019
- 2019-06-24 CN CN201910550547.1A patent/CN110321823B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20090010259A (en) * | 2007-07-20 | 2009-01-30 | 전병태 | System and method for regulating illegal headlight vehicles |
| CN102592450A (en) * | 2012-02-22 | 2012-07-18 | 杭州海康威视数字技术股份有限公司 | System and method for snapshotting behavior of motor vehicles not giving precedence to pedestrians on zebra stripes |
| CN104361747A (en) * | 2014-11-11 | 2015-02-18 | 杭州新迪数字工程系统有限公司 | Automatic capture system and recognition method for vehicles not giving way to passengers on zebra crossing |
| CN107730906A (en) * | 2017-07-11 | 2018-02-23 | 银江股份有限公司 | Zebra stripes vehicle does not give precedence to the vision detection system of pedestrian behavior |
| CN109300313A (en) * | 2017-07-25 | 2019-02-01 | 杭州海康威视数字技术股份有限公司 | Detection method, camera and the server of illegal activities |
| CN108090457A (en) * | 2017-12-27 | 2018-05-29 | 天津天地人和企业管理咨询有限公司 | A kind of motor vehicle based on video does not give precedence to pedestrian detection method |
| CN108447305A (en) * | 2018-04-21 | 2018-08-24 | 上海交通大学 | A kind of new friendship rule are lower to give precedence to pedestrian's method for early warning and early warning system |
| CN108847032A (en) * | 2018-08-21 | 2018-11-20 | 北京深瞐科技有限公司 | A kind of traffic violation recognition methods and device |
| CN109409191A (en) * | 2018-08-24 | 2019-03-01 | 广东智媒云图科技股份有限公司 | A kind of zebra stripes vehicle evacuation detection method and system based on machine learning |
| CN109035800A (en) * | 2018-09-30 | 2018-12-18 | 天津天地人和企业管理咨询有限公司 | A kind of method that intelligent transportation video camera automatic detecting machine motor-car does not avoid pedestrian |
| CN109766743A (en) * | 2018-11-21 | 2019-05-17 | 汪明智 | A kind of intelligent bionic policing system |
| CN109493609A (en) * | 2018-12-11 | 2019-03-19 | 杭州炬视科技有限公司 | A kind of portable device and method for not giving precedence to the candid photograph of pedestrian's automatic identification |
Non-Patent Citations (3)
| Title |
|---|
| TAO GAO等: "Crossing road monitoring system based on adaptive decision for illegal situation", 《APPLIED SOFT COMPUTING》 * |
| 周为钢等: "基于视频分析的机动车斑马线前未礼让违法行为检测系统", 《中国人民公安大学学报（自然科学版）》 * |
| 顾应欣: "机动车不避让行人交通违法自动检测系统研究", 《中国公共安全》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110675637A (en) * | 2019-10-15 | 2020-01-10 | 上海眼控科技股份有限公司 | Vehicle illegal video processing method and device, computer equipment and storage medium |
| CN111508082A (en) * | 2019-10-31 | 2020-08-07 | 杭州海康威视系统技术有限公司 | Parking information acquisition method and device, central platform and storage medium |
| CN111754773A (en) * | 2020-07-01 | 2020-10-09 | 宁波工程学院 | Road violation monitoring configuration method and system, camera and storage medium |
| CN112949559A (en) * | 2020-10-23 | 2021-06-11 | 深圳巴士集团股份有限公司 | Pedestrian gift detection method and device and terminal equipment |
| CN112446334A (en) * | 2020-12-02 | 2021-03-05 | 福建亿安智能技术有限公司 | Method and system for recognizing illegal behaviors of non-motor vehicle |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110321823B (en) | 2021-03-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110321823A (en) | Zebra stripes based on deep learning give precedence to the illegal secondary detection method of pedestrian | |
| CN110197589B (en) | Deep learning-based red light violation detection method | |
| CN100565555C (en) | Peccancy parking detector based on computer vision | |
| CN102521983B (en) | Vehicle violation detection system based on high definition video technology and method thereof | |
| CN103824452B (en) | A kind of peccancy parking detector based on panoramic vision of lightweight | |
| CN110136447A (en) | Lane change of driving a vehicle detects and method for distinguishing is known in illegal lane change | |
| CN109949579A (en) | A kind of illegal automatic auditing method that makes a dash across the red light based on deep learning | |
| CN106373426A (en) | Computer vision-based parking space and illegal lane occupying parking monitoring method | |
| CN106652551A (en) | Parking stall detection method and device | |
| CN107609491A (en) | A kind of vehicle peccancy parking detection method based on convolutional neural networks | |
| CN109948416A (en) | A kind of illegal occupancy bus zone automatic auditing method based on deep learning | |
| CN106448184A (en) | Identifying method of Vehicles and exit of vehicles | |
| CN106469309A (en) | The method and apparatus of vehicle monitoring, processor, image capture device | |
| KR102122859B1 (en) | Method for tracking multi target in traffic image-monitoring-system | |
| CN105185123A (en) | Fake plate vehicle recognition system and method | |
| CN107622678A (en) | A kind of intelligent traffic control system and its method based on image procossing | |
| CN110689724B (en) | Automatic motor vehicle zebra crossing present pedestrian auditing method based on deep learning | |
| CN202472943U (en) | Vehicle violation detecting system based on high definition video technology | |
| JPH05507371A (en) | Car verification system | |
| CN109741608A (en) | Motor vehicle based on deep learning, which is turned right, gives precedence to pedestrian's analysis capturing system and method | |
| CN109214345A (en) | The method for searching the driving trace for changing board vehicle is compared based on similarity | |
| CN112509325B (en) | Video deep learning-based off-site illegal automatic discrimination method | |
| CN108447305A (en) | A kind of new friendship rule are lower to give precedence to pedestrian's method for early warning and early warning system | |
| CN109671189A (en) | A kind of smart exit-entrance system based on face and vehicle linked character | |
| CN107798879A (en) | A kind of method of Intelligent Recognition fake-licensed car |
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 |