CN108875911A - One kind is parked position detecting method - Google Patents

One kind is parked position detecting method Download PDF

Info

Publication number
CN108875911A
CN108875911A CN201810516244.3A CN201810516244A CN108875911A CN 108875911 A CN108875911 A CN 108875911A CN 201810516244 A CN201810516244 A CN 201810516244A CN 108875911 A CN108875911 A CN 108875911A
Authority
CN
China
Prior art keywords
parking position
parking
point
control point
detecting method
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
Application number
CN201810516244.3A
Other languages
Chinese (zh)
Other versions
CN108875911B (en
Inventor
张�林
李曦媛
黄君豪
沈莹
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tongji University
Original Assignee
Tongji University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tongji University filed Critical Tongji University
Priority to CN201810516244.3A priority Critical patent/CN108875911B/en
Publication of CN108875911A publication Critical patent/CN108875911A/en
Application granted granted Critical
Publication of CN108875911B publication Critical patent/CN108875911B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

One kind is parked position detecting method, including：Building is extensive, band marks, towards parking position perception problems look around image data set；It is improved using the target detection frame YOLOv2 frame based on deep learning to complete the detection of parking position control point；Formation point is matched two-by-two to combination in the control point that will test, after judging that its distance restraint carries out preliminary screening, based on the shallow Model AlexNet of the disaggregated model based on depth convolutional neural networks, design self-definition model, to classifying, is completed parking position by the type and arrival line direction that judge parking position and is inferred to the parking position control point point for meeting distance restraint.The present invention is the important component of unmanned middle autonomous parking system, and the position of parking position can be only gone out by the camera detection of vehicle body surrounding, provides technical support for autonomous parking.Not only detection accuracy is high by the present invention, but also fast response time, reliable and stable.

Description

One kind is parked position detecting method
Technical field
It parks field the invention belongs to the auxiliary of computer vision and intelligent driving, is related to position detecting method of parking, especially It is the position detecting method of parking based on depth learning technology.
Background technique
Pursuit with modern humans to automation, facilitation life, all trades and professions are all towards automation, intelligentized side To development.Automobile has extremely wide market as the main vehicles of people, and automation and intelligence are also to work as One important topic of modern academia and automotive community.Automobile industry proposes the concept of intelligent vehicle, makes every effort to realize automobile It is full-automatic unmanned.Wherein, a key technology of autonomous parking system as intelligent vehicle field is increasingly becoming major One popular research direction of automobile vendor.For driver, especially beginner, limited parking space and narrow Field range makes parking become a no small challenge.
In recent years, lot of domestic and international enterprise and research institution all are putting forth effort to research and develop short distance autonomous parking system.Typically certainly The workflow of main parking system approximately as.When vehicle is close to parking area, it can be switched to the unmanned mode of low speed, And automatically along scheduled rail running.When working under unmanned mode, vehicle may need by high definition map, GPS signal or SLAM (synchronize positioning and construct map) technology to be positioned in real time.During the driving period, can search for can for vehicle Parking position, or attempt to identify and position the parking position for distributing to it by managing system of car parking.Once it is suitable to detect Parking position, vehicle will be switched to automatic parking mode, and parking path be planned, finally by vehicle parking in specified position.? In this system, the parking position Intellisense of view-based access control model is the key component in autonomous parking system.The mesh of parking position perception Be to automatically identify ground to cross the available parking position that is identified, and calculate its spatial position under vehicle axis system, To provide target position to plan for parking path.
Method early start based on detection parking position land mark is in work (Xu the et al., Vision- of Xu et al. Guided automatic parking for smart car, 2000).The method that they use image segmentation, and training For one neural network for splitting parking position graticule, its defect is can not to judge parking position type.Jung et al. Propose double touch methods (Jung et al., the Uniform user interface for for being able to detect a variety of parking positions Semiautomatic parking slot marking recognition, 2010), but such methods excessively rely on it is artificial It is selected, it is not smart enough, it is not the position detecting method of parking being fully automated.Du and Tan proposes a kind of reflexive parking system (Du et al., Autonomous reverse parking system based on robust path generation And improved sliding mode control, 2015), they use a ridge detector, are filtered by several noises Wave, the operation for removing component find out the axis of parking position label, but this method is equally not smart enough.It realizes at present fully automated Detection method (Jung et al., the Parking slot for the peak-pair that the detection method of change has Jung et al. to propose Markings recognition for automatic parking assist system, 2006), by hough space Cluster obtains the equation of warehouse compartment line, is then partitioned into the intersection point that warehouse compartment line finds out T-shape according to equation.Wang et al. proposes phase As warehouse compartment line is split in the space Radon method (Wang et al., Automatic parking based on A bird ' s eye view vision system, 2014), but both of which is only applicable to the warehouse compartment line width preset. Work (Suhr et al., the Full-automatic recognition of various parking of Suhr and Jung Slot markings using a hierarchical tree structure, 2013) it is examined using Harris Corner Detection device Angle point is measured, find out the intersection point of warehouse compartment line according to these angle points and judges parking position, this method is limited in that dependence The accuracy of Harris Corner Detection, is not sufficiently stable.
Summary of the invention
The purpose of the present invention is to provide a kind of position detecting method of parking based on deep learning assists autonomous parking system System can detect the parking position delimited by land marking line under complicated imaging circumstances, which has detection essence Spend height, fast response time, it is reliable and stable the advantages that.
In order to achieve the above object, solution of the invention is：
A kind of position detecting method of parking based on deep learning, including：Building is extensive, band marks, towards parking position Perception problems look around image data set；The detection of parking position control point is carried out using target detection frame.Based on deep learning Target detection frame has R-CNN, Fast R-CNN, Faster R-CNN, YOLO, SSD, YOLOv2.In the present invention, in order to mention High detection speed is improved using YoloV2 frame to complete the detection of parking position control point；The control point that will test is two-by-two Pairing forms point to combination, after judging that its distance restraint carries out preliminary screening, is classified using disaggregated model to it.Based on depth The disaggregated model of degree convolutional neural networks has AlexNet, GoogleNet, ResNet, ShuffleNet.In the present invention, due to Class categories number only has 7 classes, selects deep layer network that can't be obviously improved accuracy rate and takes a long time instead, therefore the present invention is with shallow Based on layer model AlexNet, design self-definition model to classifying, leads to the parking position control point point for meeting distance restraint It crosses the type for judging parking position and arrival line direction is completed parking position and inferred.
One kind is parked position detecting method, is included the following steps：
S1, the detection of parking position control point is carried out；
S2, classify to parking position arrival line；
S3, parking position deduction is carried out.
Further, parking position control point detection model described in step S1 is by the YOLOv2 mesh based on deep learning It marks detection framework and realizes that parking position arrival line disaggregated model described in step S2 is by the customized depth based on AlexNet Convolutional neural networks frame realizes that step S1, S2 is process end to end, i.e., only needs to input testing image into training gained Result can be obtained in model.
The parking position control point forms " T-type " or " L-type " angle point of parking position, in the present invention, one is parked Position includes two parking position control points and two invisible angle points.
The parking position arrival line refers to the warehouse compartment line passed through when automobile enters parking position, as connects the same pool The virtual line segment at two parking position control points of parking stall, in the present invention, parking position arrival line be it is oriented, direction is by the It is directed toward second in one parking position control point.Such as the control point point of composition parking position is to for (P1, P2), then the parking position enters Mouth line direction P2 is directed toward by P1.
Parking position control point detection model of the training based on YOLOv2, it includes following steps：
S11, prepare data, acquisition a batch looks around parking position picture construction training dataset, while manually being marked to it Note；
S12, modification network parameter, normalization constraint frame size；
After being trained on S13, data set in step s 11, parking position control point detection model is obtained.
In view of the diversity of sample, the present invention expands data using the method for equal proportion rotation, instructs to every Practice sample once to be rotated every 15 degree, data volume is made to be extended for original 24 times.When carrying out data set mark, need to remember It records the coordinate at parking position control point and can make up the control point point of parking position to serial number.
The parking position arrival line disaggregated model of the customized depth convolutional neural networks of training, it includes following steps：
S21, prepare data, concentrate marked out parking position control point to carry out two respectively the training data of step S11 Two pairings, filter out legal point pair, carry out neighborhood extraction, obtain 7 class data samples and constitute data set；
The structure and relevant parameter of S22, the customized depth convolutional neural networks of design；
S23, using the deep neural network frame designed in step S22, be trained on the data set of step S21 Afterwards, parking position arrival line disaggregated model is obtained.
Point is to what is be ordered into described in step S21, i.e., (P1, P2) and (P2, P1) are different classes of samples.In view of sample This unbalanced problem, the i.e. category distribution of training data are uneven, and the present invention uses the method for SMOTE over-sampling for data volume Less class synthesizes new samples.
The foundation of the legal point between of the screening be two parking position control points distance.
The neighborhood extraction refers to, expands △ respectively with parallel direction along a pair of legal point direction vertical to line After x and △ y pixel, image block is extracted, by making tile size for 48 × 192 pixels after scaling rotation, and point Pair line be parallel to the horizontal plane.When neighborhood has exceeded image range, then give up this to point to without processing.
After to the classification of parking position arrival line, it is inferred to reasonable parking position in the following way：
When S31, classification results are right angle parking position, " depth " of parking position is given according to priori knowledge, is inferred to residue Two do not mark the position of parking position angle point, wherein " depth " refers to the length of the other side vertical with parking position arrival line；
When S32, classification results are oblique parking position, first all angles using Gauss thread detector in inclined direction are carried out Detection, finds out the highest direction of convolution score and determines inclination angle δ, is inferred to remaining two angles further according to " depth " of parking position Point.
In terms of data set, comprising looking around image containing parking position under various complicated image-forming conditions, wherein parking position direction Including run-in index parking position, rectilinear parking position and oblique parking position；Parking position local environment includes two kinds of indoor and outdoors, Wherein outdoor environment includes that cloudy, fine day, rainy days, fine day have shade to block, are excess surface water, street lamp illumination condition, strong again Illumination condition.
7 class classification results described in parking position arrival line disaggregated model are respectively：(a) right angle of parking position direction upward Parking position；(b) oblique parking position of the parking position direction towards upper right；(c) oblique parking position of the parking position direction towards upper left；(d) it moors The right angle parking position that parking stall is directed downward；(e) oblique parking position of the parking position direction towards lower-left；(f) parking position direction is towards bottom right Oblique parking position；(g) parking position is not constituted.
The input picture size of the customized depth convolutional neural networks frame is 48 × 192 pixels, and output layer has 7 A node respectively corresponds 7 class classification results of model output.Customized depth convolutional neural networks frame includes 4 convolution Layer, 3 maximum pond layers, 2 normalization layers and 2 full articulamentums.
By adopting the above-described technical solution, the invention has the advantages that：It is able to solve under complicated image-forming condition Parking position test problems, and detection accuracy is high, fast response time, reliable and stable.
Detailed description of the invention
Fig. 1 (a), Fig. 1 (b), Fig. 1 (c) show heretofore described parking position control point, arrival line and separator bar, Position view in parallel space, vertical parking stall and oblique parking stall respectively, wherein 1 is parking position control point, 2 be parking position Arrival line, 3 be parking position separator bar.
Fig. 2 show it is heretofore described in looking around image by two parking position control point P1And P2Composed pool Topography's block of parking stall arrival line.
It is respectively institute in the present invention shown in Fig. 3 (a), Fig. 3 (b), Fig. 3 (c), Fig. 3 (d), Fig. 3 (e), Fig. 3 (f), Fig. 3 (g) The 7 class parking position arrival line schematic diagrames stated.
Fig. 4 show the frame of the heretofore described customized convolutional neural networks disaggregated model based on AlexNet Structural schematic diagram.
Fig. 5 (a), Fig. 5 (b) show heretofore described right angle parking position and infer schematic diagram.
Fig. 6 (a), Fig. 6 (b) show heretofore described oblique parking position and infer schematic diagram.
Fig. 7 show heretofore described General Implementing step schematic diagram.
Specific embodiment
The present invention is further illustrated below in conjunction with the embodiment shown in that figure.
Position detecting method of parking of the invention is the important component of autonomous parking system, can only pass through vehicle body surrounding Camera detection go out the location information of parking position, provide technical support for decision rule layer.
Design of the invention can meet the needs for detecting vertical, parallel and oblique parking stall simultaneously.In parking stall measure on the berth, The most apparent feature of parking position of the ground scribing line of view-based access control model is parking position control point, Fig. 1 (a), Fig. 1 (b), Fig. 1 (c) institute It is shown as the parking position control point of each type, the schematic diagram of arrival line and separator bar, as seen from the figure, parking position arrival line is by one To parking position control point composition.The present invention acquires the pool of the condition of satisfaction by detection T-shape or " L " type parking position control point Parking stall arrival line, then by classifying to it, determine the direction of arrival line, carry out parking stall deduction.This detection method is divided into three ranks Section, the first step detect parking position control point, and second step divides arrival line composed by the parking position control point detected Class, third step infer parking position position according to classification results.Detailed process is：
One, parking position control point is detected
Parking position control point forms " T-type " or " L-type " angle point of parking position, refer to parking position arrival line and divide Local segment centered on line crosspoint, as shown in Fig. 1 (a), Fig. 1 (b), Fig. 1 (c), circles mark has gone out institute in the present invention Position of the parking position control point stated in parking position.In the present invention, parking position include two parking position control points and Two invisible angle points.
In terms of data set, looking around image data and establish parking position data set with acquisition.Comprising various complexity at slice Image is looked around containing parking position under part, wherein parking position direction includes run-in index parking position, rectilinear parking position and oblique Parking position；Parking position local environment includes two kinds of indoor and outdoors, wherein outdoor environment again include cloudy, fine day, it is rainy days, fine It has shade to block, excess surface water, street lamp illumination condition, strong illumination condition.In the present invention, the panoramic view of data set is constituted The resolution ratio of picture is 416 × 416 pixels.
The present invention carries out the detection of parking position control point using the YOLOv2 target detection frame based on deep learning.
Before training detects network, needs to prepare training sample, on the parking position data set of building, mark out by hand The position at all parking position control points.For each control point Pi, with PiCentered on fixed size p × p square-shaped frame quilt It is considered PiGround-truth (correct data).Access is according to the image with mark of concentration 4/5ths as parking position The training sample of control point detection model, the image of residue 1/5th is as test set, for testing the detection of assessment models Performance.
In parking stall measure on the berth, characteristic that trained parking position control point detection model needs to have invariable rotary. In order to achieve this goal, the present invention generates the sample of some rotations by rotating each original mark image to expand instruction Practice collection.Specifically, for each original mark image I, its available J rotated sampleEach IjIt is rotated by image IDegree generates.Likewise, the mark coordinate at parking position control point is also done in the same fashion rotation. It follows that there will be JN image patterns if being labelled with N training sample images to train parking position control point to detect mould Type.
In terms of network parameter setting, the constraint frame preset in YOLOv2 frame has 5 kinds of sizes, training sample in the present invention Ground-truth only have a kind of and be fixed size, be trained using the constraint frame of excessive size will lead to mistake instead Difference is excessive and makes the reduction of model convergence rate.Therefore in the present invention, the constraint frame type and size of YOLOv2 are modified, Constraint frame type is changed to one kind, is the image of M × N pixel for training sample size, default length and width are respectivelyWith
After completing above-mentioned work, model training is carried out offline, and a model D is trained to carry out the detection of parking position control point.
In detection-phase, it is only necessary to testing image is input in the resulting model of training, can be obtained as a result, exporting inspection The position coordinates at the parking position control point measured, this is a process end to end.
Two, parking position arrival line is classified
After being detected by parking position control point, K parking position control point is obtained, they are combined to pairing two-by-two.If P1And P2It is a pair of point pair composed by control point that two of them are detected, then needs to verify P1And P2Whether it is capable of forming Effective parking position arrival line.It indicates from P1It is directed toward P2Directed line segment, | | P1P2| | indicate P1And P2The distance between.
1, for effective parking position arrival line candidate item, P1And P2The distance between should meet following constraint condition.
1-It is the arrival line candidate item of a Parallel parking position, it needs to meet t1<||P1P2||<t2；
1-2It is the arrival line candidate item of a vertical or oblique parking position, it needs to meet t3<||P1P2||<t4；
Wherein, t1,t2,t3,t4Size be to be arranged according to the priori knowledge of the entrance line length of different type parking position Distance parameter.
According to step 1-1 and 1-2, preliminary screening goes out to meet the parking position arrival line candidate item of constraint condition.
2, it setsThe parking position arrival line candidate item picked out by step 1, need to carry out it classification judge into Mouth line direction and parking position type, detailed process are：
2-1 pairsCarry out neighborhood extraction.Fig. 2 is shown in looking around image by two parking position control point P1And P2Group At parking position arrival line topography's block.A local coordinate system is established, accordingly with P1And P2Center be origin, withFor x-axis, the y-axis vertical with x-axis can then be uniquely determined.In this coordinate system, a rectangular area R is defined, it is closed It is all symmetrical in x-axis and y-axis.For region R, it is set as along the side length of x-axis | | P1P2| |+Δ x, along the side of y-axis A length of Δ y.It is to extract region R that the neighborhood, which extracts, and region R is extracted in image from looking around, and transformation is carried out to it keeps x-axis parallel It is normalized in horizontal plane, and to its size as w × h pixel, preferred embodiment w=48, h=192.
The parking position arrival line disaggregated model of customized depth convolutional neural networks of the 2-2 training based on AlexNet.
In terms of data set, training set of images is looked around based on what the band in step 1 marked, an available set C, it Contain all neighborhood R defined by two parking position control points, by C point be 7 classes according to the type of parking position, as Fig. 3 (a), Shown in Fig. 3 (b), Fig. 3 (c), Fig. 3 (d), Fig. 3 (e), Fig. 3 (f), Fig. 3 (g), this 7 class is respectively：Fig. 3 (a) parking position direction court On right angle parking position；Oblique parking position of Fig. 3 (b) parking position direction towards upper right；Fig. 3 (c) parking position direction is towards the oblique of upper left To parking position；The right angle parking position that Fig. 3 (d) parking position is directed downward；Oblique parking position of Fig. 3 (e) parking position direction towards lower-left； Oblique parking position of Fig. 3 (f) parking position direction towards bottom right；Fig. 3 (g) does not constitute parking position.When constructing C, in order to solve class not The problem of balance, the i.e. sample of some particular categories, are less relative to other categorical measures, and present invention employs SMOTE The method of (Synthetic Minority Over-sampling Technique) adopted the classification of negligible amounts Sample increases its sample size.In present invention, it is desirable to converting the color image of training sample to gray level image.
In the training stage, a disaggregated model M is trained to predict from the classification for looking around the neighborhood R extracted in image.For Disaggregated model M, the present invention in using based on AlexNet customized depth convolutional neural networks frame realize, be illustrated in figure 4 The frame structure of disaggregated model, the input of network are 48 × 192 gray level images, and output layer has 7 nodes, are respectively corresponded described 7 class neighborhood R.Conv indicates that a convolutional layer, ReLU indicate an amendment linear unit in Fig. 4, and max-pool expression one is most Great Chiization layer, LRN indicate that a Local Phase should normalize layer, and FC indicates that a full articulamentum, dropout indicate a discarding Layer.For each layer, other than ReLU, FC and dropout layers, used parameter is as shown in table 1：
Table 1
For LRN layers, useIt indicates at position (x, y), after being calculated using core (kernel) i, through overdriving
Output afterwards, then after LRN layers, next layer of input of normalized responseFor：
Wherein, on identical spatial position, N is the sum of the kernel of this layer, and n expression takes centered on the kernel Left and right each n/2 kernel average.
After completing above-mentioned work, model training is carried out offline, obtains parking position arrival line disaggregated model M.
In detection-phase, it is only necessary to testing image is inputed into the resulting model of training, can output category result, this is One process end to end.
2-3 is inputted parking position arrival line model M, obtains for neighborhood R described in step 2-1Classification As a result.
Three, parking position is inferred
In parking stall measure on the berth, parking position is a parallelogram, by the coordinate representation on four vertex.In the present invention In, the angle point at two non-mark control points is not in field range, they can only be obtained by reasoning.According to priori knowledge The depth (i.e. the length of separator bar) of given parking position, the depth of vertical, parallel and oblique parking stall is respectively d1,d2,d3。3-1P1 And P2The respectively two parking position control points being detected, by P1And P2The neighborhood R of extraction is classified as right angle parking position, this When remaining two non-controlling points angle point P3And P4It can be easy to calculate, as shown in Fig. 5 (a), right angles parking stall is located atClockwise direction, just like giving a definition：
As shown in Fig. 5 (b), right angle parallel space is located atCounter clockwise direction, just like giving a definition：
3-2P1And P2The respectively two parking position control points being detected, by P1And P2The neighborhood R of extraction is classified as Oblique parking position, at this time the angle point P of remaining two non-controlling points3And P4It then also needs an inclination angle δ that can just calculate, is Solution this problem, the present invention use Gauss thread detector based on template, carry out it in all angles of inclined direction Detection, finds out the highest direction of convolution score to determine the value of δ.The specific method is as follows：
As shown in Fig. 6 (a), prepare one group of ideal " T-type " template offlineWherein θjIt is two of template j straight The angle of line, L are the sums of template.The size of each template is s × s, and carries out zero averaging to each template.In test rank Section, with P1And P2Centered on extract the image block I of s × s size respectively1And I2, and I1And I2Be aboutSymmetrically, then mooring Parking stall inclination angle δ can be defined as：Wherein, * indicates related operation.
After calculating inclination angle δ, the angle point P of available remaining two non-controlling point3And P4.As shown in Fig. 6 (b), to the right Inclined oblique parking stall is located atCounter clockwise direction, just like giving a definition：
By above step, the completely position detecting method of parking based on deep learning, General Implementing step are established It is rapid as shown in Figure 7.
Beneficial effects of the present invention are illustrated below in conjunction with specific experiment：
Experimental setup：In order to which acquire certain scale looks around parking position data, more of the invention can reach for analyzing Effect, in an experiment, using Roewe E50 electric car carry out data acquisition, acquire 12165 in total and look around image, In 9827 be used as training set, 2338 be used as test set.Experiment is being furnished with 2.4GHZ Intel Xeon E5-2630V3CPU, It is run on the work station of Nvidia Titan X video card and 32GB RAM, programming language C++.
Experiment one：Detect evaluation experimental in parking position control point：
In the present invention, the detection of parking position control point is a crucial step.In order to objectively evaluate detection of the invention Can, comparative experiments is set, on the data set that the present invention constructs, realizes the algorithm of target detection of some classics.
The contrast method of reference includes：Method " the Robust that P.A.Viola and M.J.Jones was proposed in 2004 Real-time face detection ", is denoted as " VJ " in an experiment；What N.Dalal and B.Triggs was proposed in 2005 Method " Histogram of oriented gradients for human detection ", is denoted as " HoG+ in an experiment SVM"；Method " the An HOG-LBP human detector with that X.Wang, X.Han, and S.Yan were proposed in 2009 Partial occlusion handling ", is denoted as " HoG+LBP " in an experiment；W.Schwarts,A.Kembhavi, Method " the Human detection using partial least that D.Harwood, and L.Davis were proposed in 2009 Squares analysis " is denoted as " PLS " in an experiment；" the A that C.Wojek and B.Schiele was proposed in 2008 Performance evaluation of single and multifeature people detection ", in an experiment It is denoted as " MultiFtr "；What R.Benenson, M.Mathias, T.Tuytelaars, and L.Van Gool were proposed in 2013 years " Seeking the strongest rigid detector ", is denoted as " Roerei " in an experiment；P.Dollar,C.Wojek, " the Pedestrian detection that B.Schiele, and P.Perona were proposed in 2012:An evaluation of The state of the art ", is denoted as " ACF+Boosting " in an experiment.
Appreciation gist is to calculate average log miss rate (fppi) and position error, and every control side is summarized in table 2 The method that its is proposed looks around the experimental result on data set in parking position in method and the present invention.
Table 2
As shown in Table 2, parking position according to the present invention controls point detecting method, and detection accuracy is far longer than other classics Algorithm.
Experiment two：Parking position detects evaluation experimental
In order to detect this method finally to the accuracy of identification of parking stall, using following evaluation index：
Summarized in table 3 proposed in the present invention based on deep learning park position detecting method on test set not With the performance situation of the data sample under environment.
Table 3
According to the experimental results, the position detecting method of parking proposed by the present invention based on deep learning is in various varying environments Under the conditions of, performance is stablized, and detection accuracy is high.
In terms of calculating speed, the present invention is in 2.4GHZ Intel Xeon E5-2630V3CPU, Nvidia Titan X The efficiency of 43fps can be reached on the work station of video card and 32GB RAM.The present invention can achieve 10fps on vehicle-mounted TX2 platform Efficiency, while verification and measurement ratio can be taken into account.The present invention is demonstrated to use under actual scene, is one in autonomous parking field Initiative method, can be used as comparison pedestal method.
This hair can be understood and applied the above description of the embodiments is intended to facilitate those skilled in the art It is bright.Person skilled in the art obviously easily can make various modifications to these embodiments, and described herein General Principle is applied in other embodiments without having to go through creative labor.Therefore, the present invention is not limited to the above embodiments, Those skilled in the art's announcement according to the present invention, improvement and modification made without departing from the scope of the present invention all should be in this hairs Within bright protection scope.

Claims (10)

  1. The position detecting method 1. one kind is parked, which is characterized in that include the following steps：
    S1, the detection of parking position control point is carried out using target detection frame；
    S2, classified using disaggregated model to parking position arrival line；
    S3, parking position deduction is carried out.
  2. 2. position detecting method according to claim 1 of parking, which is characterized in that including：Building is extensive, band marks, Image data set is looked around towards parking position perception problems；Formation point is matched two-by-two to combination, to upper in the control point that will test State a little to carry out preliminary screening after, classified using the self-definition model based on depth convolutional neural networks to it, by sentencing The type of disconnected parking position and arrival line direction are completed parking position and are inferred.
  3. 3. position detecting method according to claim 1 of parking, which is characterized in that target detection frame described in step S1 is Target detection frame based on deep learning, including R-CNN, Fast R-CNN, Faster R-CNN, YOLO, SSD, YOLOv2 Target detection frame, preferably Yolov2 target detection frame；
    Preferably, disaggregated model described in step S2 is the disaggregated model based on depth convolutional neural networks, including AlexNet, GoogleNet, ResNet, ShuffleNet are preferably based on the customized depth convolutional neural networks frame of AlexNet.
  4. 4. position detecting method according to claim 3 of parking, which is characterized in that parking position control of the training based on YOLOv2 Point detection model, it includes following steps：
    S11, prepare data, acquisition a batch looks around parking position picture construction training dataset, while manually being marked to it；
    S12, modification network parameter, normalization constraint frame size；
    After being trained on S13, data set in step s 11, parking position control point detection model is obtained；
    Preferably, it is contemplated that sample diversity expands data using the method that equal proportion rotates, to every training sample It is once rotated every 15 degree, data volume is made to be extended for original 24 times；
    Preferably, it needs to record the coordinate at parking position control point when data set marks and can make up the control point point pair of parking position Serial number.
  5. 5. position detecting method according to claim 4 of parking, which is characterized in that the customized depth convolutional neural networks of training Parking position arrival line disaggregated model, it includes following steps：
    S21, prepare data, concentrate marked out parking position control point to be matched two-by-two respectively the training data of step S11 It is right, legal point pair is filtered out, neighborhood extraction is carried out, 7 class data samples is obtained and constitutes data set；
    S22, the customized depth convolutional neural networks structure of design and relevant parameter；
    S23, it is obtained after being trained on the data set of step S21 using the deep neural network frame designed in step S22 To parking position arrival line disaggregated model；
    Preferably, point is to what is be ordered into described in step S21, i.e., (P1, P2) and (P2, P1) are different classes of samples；
    Preferably, it is contemplated that the unbalanced problem of sample, the i.e. category distribution of training data are uneven, using SMOTE over-sampling Method be that the less class of data volume synthesizes new samples.
  6. 6. position detecting method according to claim 1 of parking, which is characterized in that classify to parking position arrival line Afterwards, it is inferred to reasonable parking position in the following way：
    When S31, classification results are right angle parking position, " depth " of parking position is given according to priori knowledge, is inferred to two remaining The position of parking position angle point is not marked, wherein " depth " refers to the length of the other side vertical with parking position arrival line；
    When S32, classification results are oblique parking position, first all angles using Gauss thread detector in inclined direction are visited It surveys, finds out the highest direction of convolution score and determine inclination angle δ, be inferred to remaining angle point further according to " depth " of parking position.
  7. 7. position detecting method according to claim 4 of parking, which is characterized in that detect mould in the parking position control point Type, data set aspect, parking position direction includes run-in index parking position, rectilinear parking position and oblique parking position；Parking position institute Locating environment includes two kinds of indoor and outdoors, wherein outdoor environment include again cloudy, fine day, rainy days, fine day there is shade to block, Area water, street lamp illumination condition, strong illumination condition.
  8. 8. position detecting method according to claim 5 of parking, which is characterized in that the parking position arrival line classification mould Type, in step S21, the foundation of the legal point between of screening be two parking position control points distance；
    Preferably, in step S21, neighborhood extraction refers to, along a pair of legal point direction vertical to line with it is parallel After △ x and △ y pixel is expanded in direction respectively, extract image block, by scaling rotation after make tile size be 48 × 192 pixels, and the line of point pair is parallel to the horizontal plane；When neighborhood has exceeded image range, then give up this to point pair Without processing.
  9. 9. position detecting method according to claim 5 of parking, which is characterized in that the parking position arrival line classification mould Type, the 7 class data samples are respectively：(a) the right angle parking position of parking position direction upward；(b) parking position direction is towards upper right Oblique parking position；(c) oblique parking position of the parking position direction towards upper left；(d) the right angle parking position that parking position is directed downward； (e) oblique parking position of the parking position direction towards lower-left；(f) oblique parking position of the parking position direction towards bottom right；(g) it does not constitute and parks Position.
  10. 10. position detecting method according to claim 5 of parking, which is characterized in that the parking position arrival line classification mould Type, in step S22, the input picture size of the customized depth convolutional neural networks frame is 48 × 192 pixels, output Layer has 7 nodes, respectively corresponds 7 class classification results described in claim 7；
    Preferably, customized depth convolutional neural networks frame includes 4 convolutional layers, 3 maximum pond layers, 2 normalization layers With 2 full articulamentums.
CN201810516244.3A 2018-05-25 2018-05-25 Parking space detection method Active CN108875911B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810516244.3A CN108875911B (en) 2018-05-25 2018-05-25 Parking space detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810516244.3A CN108875911B (en) 2018-05-25 2018-05-25 Parking space detection method

Publications (2)

Publication Number Publication Date
CN108875911A true CN108875911A (en) 2018-11-23
CN108875911B CN108875911B (en) 2021-06-18

Family

ID=64333891

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810516244.3A Active CN108875911B (en) 2018-05-25 2018-05-25 Parking space detection method

Country Status (1)

Country Link
CN (1) CN108875911B (en)

Cited By (38)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109583392A (en) * 2018-12-05 2019-04-05 北京纵目安驰智能科技有限公司 A kind of method for detecting parking stalls, device and storage medium
CN109614914A (en) * 2018-12-05 2019-04-12 北京纵目安驰智能科技有限公司 Parking stall vertex localization method, device and storage medium
CN109614913A (en) * 2018-12-05 2019-04-12 北京纵目安驰智能科技有限公司 A kind of oblique parking stall recognition methods, device and storage medium
CN109635750A (en) * 2018-12-14 2019-04-16 广西师范大学 A kind of compound convolutional neural networks images of gestures recognition methods under complex background
CN109800658A (en) * 2018-12-26 2019-05-24 中汽研（天津）汽车工程研究院有限公司 Parking position type online recognition neural network based and positioning system and method
CN109977862A (en) * 2019-03-26 2019-07-05 北京茵沃汽车科技有限公司 A kind of recognition methods of parking stall limiter
CN109977817A (en) * 2019-03-14 2019-07-05 南京邮电大学 EMU car bed bolt fault detection method based on deep learning
CN110132278A (en) * 2019-05-14 2019-08-16 驭势科技（北京）有限公司 A kind of instant method and device for positioning and building figure
CN110348407A (en) * 2019-07-16 2019-10-18 同济大学 One kind is parked position detecting method
CN110348297A (en) * 2019-05-31 2019-10-18 纵目科技（上海）股份有限公司 A kind of detection method, system, terminal and the storage medium of parking systems for identification
CN110706509A (en) * 2019-10-12 2020-01-17 东软睿驰汽车技术（沈阳）有限公司 Parking space and direction angle detection method, device, equipment and medium thereof
CN110901632A (en) * 2019-11-29 2020-03-24 长城汽车股份有限公司 Automatic parking control method and device
CN111178236A (en) * 2019-12-27 2020-05-19 清华大学苏州汽车研究院（吴江） Parking space detection method based on deep learning
CN111260955A (en) * 2018-12-03 2020-06-09 初速度（苏州）科技有限公司 Parking space detection system and method adopting parking space frame lines and end points
CN111376895A (en) * 2018-12-29 2020-07-07 上海汽车集团股份有限公司 Around-looking parking sensing method and device, automatic parking system and vehicle
CN111580131A (en) * 2020-04-08 2020-08-25 西安邮电大学 Method for identifying vehicles on expressway by three-dimensional laser radar intelligent vehicle
US20200284914A1 (en) * 2019-03-05 2020-09-10 Visteon Global Technologies, Inc. Multiple vertical layer light detection and ranging system, auto-parking assistance, and computer vision lane detection and keeping
CN112052782A (en) * 2020-08-31 2020-12-08 安徽江淮汽车集团股份有限公司 Around-looking-based parking space identification method, device, equipment and storage medium
CN112130146A (en) * 2020-08-26 2020-12-25 南京航空航天大学 Video synthetic aperture radar moving target bright line detection method based on Radon transformation and machine learning
CN112172798A (en) * 2020-09-28 2021-01-05 惠州华阳通用电子有限公司 Parking method based on water accumulation environment and storage medium
CN112193240A (en) * 2020-09-28 2021-01-08 惠州华阳通用电子有限公司 Parking method based on water accumulation information
CN112308922A (en) * 2020-11-25 2021-02-02 南京鸿昌智能货架有限公司 Deep learning-based automatic goods-searching positioning method and system for shuttle
CN112356826A (en) * 2020-10-28 2021-02-12 惠州华阳通用电子有限公司 Parking assisting method and storage medium
CN112417926A (en) * 2019-08-22 2021-02-26 广州汽车集团股份有限公司 Parking space identification method and device, computer equipment and readable storage medium
CN112598922A (en) * 2020-12-07 2021-04-02 安徽江淮汽车集团股份有限公司 Parking space detection method, device, equipment and storage medium
CN112669615A (en) * 2020-12-09 2021-04-16 上汽大众汽车有限公司 Parking space detection method and system based on camera
CN112668588A (en) * 2020-12-29 2021-04-16 禾多科技（北京）有限公司 Parking space information generation method, device, equipment and computer readable medium
CN113011355A (en) * 2021-03-25 2021-06-22 东北林业大学 Pine wood nematode disease image recognition detection method and device
CN113240734A (en) * 2021-06-01 2021-08-10 深圳市捷顺科技实业股份有限公司 Vehicle straddle judgment method, device, equipment and medium based on aerial view
CN113320474A (en) * 2021-07-08 2021-08-31 长沙立中汽车设计开发股份有限公司 Automatic parking method and device based on panoramic image and human-computer interaction
CN113627276A (en) * 2021-07-20 2021-11-09 的卢技术有限公司 Method and device for detecting parking space
CN113627277A (en) * 2021-07-20 2021-11-09 的卢技术有限公司 Method and device for identifying parking space
WO2022016355A1 (en) * 2020-07-21 2022-01-27 华为技术有限公司 Method and apparatus for determining labeling information
CN114155740A (en) * 2021-12-20 2022-03-08 上海高德威智能交通系统有限公司 Parking space detection method, device and equipment
CN115601971A (en) * 2022-11-12 2023-01-13 广州融嘉信息科技有限公司（Cn） Park self-adaptive vehicle scheduling and parking intelligent control method based on neural network
CN116310390A (en) * 2023-05-17 2023-06-23 上海仙工智能科技有限公司 Visual detection method and system for hollow target and warehouse management system
RU2800036C1 (en) * 2019-11-29 2023-07-17 Грейт Волл Мотор Компани Лимитед Method for controlling automated parking system and devices for its implementation
EP4296143A3 (en) * 2022-06-01 2024-04-17 Anhui NIO Autonomous Driving Technology Co., Ltd. Method for determining parking space entrance line, computer device, storage medium, and vehicle

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050195383A1 (en) * 1994-05-23 2005-09-08 Breed David S. Method for obtaining information about objects in a vehicular blind spot
US20080119993A1 (en) * 1994-05-23 2008-05-22 Automotive Technologies International, Inc. Exterior Airbag Deployment Techniques
CN102663357A (en) * 2012-03-28 2012-09-12 北京工业大学 Color characteristic-based detection algorithm for stall at parking lot
CN105975941A (en) * 2016-05-31 2016-09-28 电子科技大学 Multidirectional vehicle model detection recognition system based on deep learning
CN107886080A (en) * 2017-11-23 2018-04-06 同济大学 One kind is parked position detecting method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050195383A1 (en) * 1994-05-23 2005-09-08 Breed David S. Method for obtaining information about objects in a vehicular blind spot
US20080119993A1 (en) * 1994-05-23 2008-05-22 Automotive Technologies International, Inc. Exterior Airbag Deployment Techniques
CN102663357A (en) * 2012-03-28 2012-09-12 北京工业大学 Color characteristic-based detection algorithm for stall at parking lot
CN105975941A (en) * 2016-05-31 2016-09-28 电子科技大学 Multidirectional vehicle model detection recognition system based on deep learning
CN107886080A (en) * 2017-11-23 2018-04-06 同济大学 One kind is parked position detecting method

Cited By (59)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111260955A (en) * 2018-12-03 2020-06-09 初速度（苏州）科技有限公司 Parking space detection system and method adopting parking space frame lines and end points
CN109614914A (en) * 2018-12-05 2019-04-12 北京纵目安驰智能科技有限公司 Parking stall vertex localization method, device and storage medium
CN109614913A (en) * 2018-12-05 2019-04-12 北京纵目安驰智能科技有限公司 A kind of oblique parking stall recognition methods, device and storage medium
CN109583392A (en) * 2018-12-05 2019-04-05 北京纵目安驰智能科技有限公司 A kind of method for detecting parking stalls, device and storage medium
CN109635750A (en) * 2018-12-14 2019-04-16 广西师范大学 A kind of compound convolutional neural networks images of gestures recognition methods under complex background
CN109800658A (en) * 2018-12-26 2019-05-24 中汽研（天津）汽车工程研究院有限公司 Parking position type online recognition neural network based and positioning system and method
CN111376895A (en) * 2018-12-29 2020-07-07 上海汽车集团股份有限公司 Around-looking parking sensing method and device, automatic parking system and vehicle
US20200284914A1 (en) * 2019-03-05 2020-09-10 Visteon Global Technologies, Inc. Multiple vertical layer light detection and ranging system, auto-parking assistance, and computer vision lane detection and keeping
CN109977817B (en) * 2019-03-14 2021-04-27 南京邮电大学 Motor train unit bottom plate bolt fault detection method based on deep learning
CN109977817A (en) * 2019-03-14 2019-07-05 南京邮电大学 EMU car bed bolt fault detection method based on deep learning
CN109977862B (en) * 2019-03-26 2021-10-15 北京茵沃汽车科技有限公司 Recognition method of parking space limiter
CN109977862A (en) * 2019-03-26 2019-07-05 北京茵沃汽车科技有限公司 A kind of recognition methods of parking stall limiter
CN110132278A (en) * 2019-05-14 2019-08-16 驭势科技（北京）有限公司 A kind of instant method and device for positioning and building figure
CN110348297B (en) * 2019-05-31 2023-12-26 纵目科技（上海）股份有限公司 Detection method, system, terminal and storage medium for identifying stereo garage
CN110348297A (en) * 2019-05-31 2019-10-18 纵目科技（上海）股份有限公司 A kind of detection method, system, terminal and the storage medium of parking systems for identification
CN110348407A (en) * 2019-07-16 2019-10-18 同济大学 One kind is parked position detecting method
CN112417926B (en) * 2019-08-22 2024-02-27 广州汽车集团股份有限公司 Parking space identification method and device, computer equipment and readable storage medium
CN112417926A (en) * 2019-08-22 2021-02-26 广州汽车集团股份有限公司 Parking space identification method and device, computer equipment and readable storage medium
CN110706509A (en) * 2019-10-12 2020-01-17 东软睿驰汽车技术（沈阳）有限公司 Parking space and direction angle detection method, device, equipment and medium thereof
WO2021068588A1 (en) * 2019-10-12 2021-04-15 东软睿驰汽车技术（沈阳）有限公司 Method and apparatus for detecting parking space and direction and angle thereof, device and medium
WO2021104476A1 (en) * 2019-11-29 2021-06-03 长城汽车股份有限公司 Automatic parking control method and apparatus
RU2800036C1 (en) * 2019-11-29 2023-07-17 Грейт Волл Мотор Компани Лимитед Method for controlling automated parking system and devices for its implementation
CN110901632A (en) * 2019-11-29 2020-03-24 长城汽车股份有限公司 Automatic parking control method and device
EP4023517A4 (en) * 2019-11-29 2022-11-09 Great Wall Motor Company Limited Automatic parking control method and apparatus
JP7379683B2 (en) 2019-11-29 2023-11-14 長城汽車股▲分▼有限公司 Automatic parking control method and device
JP2022551272A (en) * 2019-11-29 2022-12-08 長城汽車股▲分▼有限公司 Automatic parking control method and device
US11745730B2 (en) 2019-11-29 2023-09-05 Great Wall Motor Company Limited Automatic parking control method and apparatus
CN111178236B (en) * 2019-12-27 2023-06-06 清华大学苏州汽车研究院（吴江） Parking space detection method based on deep learning
CN111178236A (en) * 2019-12-27 2020-05-19 清华大学苏州汽车研究院（吴江） Parking space detection method based on deep learning
CN111580131B (en) * 2020-04-08 2023-07-07 西安邮电大学 Method for identifying vehicles on expressway by three-dimensional laser radar intelligent vehicle
CN111580131A (en) * 2020-04-08 2020-08-25 西安邮电大学 Method for identifying vehicles on expressway by three-dimensional laser radar intelligent vehicle
WO2022016355A1 (en) * 2020-07-21 2022-01-27 华为技术有限公司 Method and apparatus for determining labeling information
CN112130146A (en) * 2020-08-26 2020-12-25 南京航空航天大学 Video synthetic aperture radar moving target bright line detection method based on Radon transformation and machine learning
CN112130146B (en) * 2020-08-26 2022-05-03 南京航空航天大学 Video synthetic aperture radar moving target bright line detection method based on Radon transformation and machine learning
CN112052782A (en) * 2020-08-31 2020-12-08 安徽江淮汽车集团股份有限公司 Around-looking-based parking space identification method, device, equipment and storage medium
CN112052782B (en) * 2020-08-31 2023-09-05 安徽江淮汽车集团股份有限公司 Method, device, equipment and storage medium for recognizing parking space based on looking around
CN112193240B (en) * 2020-09-28 2022-02-01 惠州华阳通用电子有限公司 Parking method based on water accumulation information
CN112172798B (en) * 2020-09-28 2022-02-01 惠州华阳通用电子有限公司 Parking method based on water accumulation environment and storage medium
CN112193240A (en) * 2020-09-28 2021-01-08 惠州华阳通用电子有限公司 Parking method based on water accumulation information
CN112172798A (en) * 2020-09-28 2021-01-05 惠州华阳通用电子有限公司 Parking method based on water accumulation environment and storage medium
CN112356826B (en) * 2020-10-28 2022-02-01 惠州华阳通用电子有限公司 Parking assisting method and storage medium
CN112356826A (en) * 2020-10-28 2021-02-12 惠州华阳通用电子有限公司 Parking assisting method and storage medium
CN112308922A (en) * 2020-11-25 2021-02-02 南京鸿昌智能货架有限公司 Deep learning-based automatic goods-searching positioning method and system for shuttle
CN112598922A (en) * 2020-12-07 2021-04-02 安徽江淮汽车集团股份有限公司 Parking space detection method, device, equipment and storage medium
CN112669615A (en) * 2020-12-09 2021-04-16 上汽大众汽车有限公司 Parking space detection method and system based on camera
CN112668588B (en) * 2020-12-29 2023-09-12 禾多科技（北京）有限公司 Parking space information generation method, device, equipment and computer readable medium
CN112668588A (en) * 2020-12-29 2021-04-16 禾多科技（北京）有限公司 Parking space information generation method, device, equipment and computer readable medium
CN113011355A (en) * 2021-03-25 2021-06-22 东北林业大学 Pine wood nematode disease image recognition detection method and device
CN113240734A (en) * 2021-06-01 2021-08-10 深圳市捷顺科技实业股份有限公司 Vehicle straddle judgment method, device, equipment and medium based on aerial view
CN113240734B (en) * 2021-06-01 2024-05-17 深圳市捷顺科技实业股份有限公司 Vehicle cross-position judging method, device, equipment and medium based on aerial view
CN113320474A (en) * 2021-07-08 2021-08-31 长沙立中汽车设计开发股份有限公司 Automatic parking method and device based on panoramic image and human-computer interaction
CN113627276A (en) * 2021-07-20 2021-11-09 的卢技术有限公司 Method and device for detecting parking space
CN113627277A (en) * 2021-07-20 2021-11-09 的卢技术有限公司 Method and device for identifying parking space
CN114155740A (en) * 2021-12-20 2022-03-08 上海高德威智能交通系统有限公司 Parking space detection method, device and equipment
EP4296143A3 (en) * 2022-06-01 2024-04-17 Anhui NIO Autonomous Driving Technology Co., Ltd. Method for determining parking space entrance line, computer device, storage medium, and vehicle
CN115601971A (en) * 2022-11-12 2023-01-13 广州融嘉信息科技有限公司（Cn） Park self-adaptive vehicle scheduling and parking intelligent control method based on neural network
CN115601971B (en) * 2022-11-12 2023-11-10 广州融嘉信息科技有限公司 Park self-adaptive vehicle dispatching and parking intelligent control method based on neural network
CN116310390A (en) * 2023-05-17 2023-06-23 上海仙工智能科技有限公司 Visual detection method and system for hollow target and warehouse management system
CN116310390B (en) * 2023-05-17 2023-08-18 上海仙工智能科技有限公司 Visual detection method and system for hollow target and warehouse management system

Also Published As

Publication number Publication date
CN108875911B (en) 2021-06-18

Similar Documents

Publication Publication Date Title
CN108875911A (en) One kind is parked position detecting method
CN106250870B (en) A kind of pedestrian&#39;s recognition methods again of joint part and global similarity measurement study
CN106127204B (en) A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks
CN106225787B (en) Unmanned aerial vehicle visual positioning method
CN107886080A (en) One kind is parked position detecting method
Siagian et al. Biologically inspired mobile robot vision localization
CN109635875A (en) A kind of end-to-end network interface detection method based on deep learning
CN110378376A (en) A kind of oil filler object recognition and detection method based on machine vision
CN109460709A (en) The method of RTG dysopia analyte detection based on the fusion of RGB and D information
CN109711288A (en) Remote sensing ship detecting method based on feature pyramid and distance restraint FCN
Miao et al. On-board lane detection system for intelligent vehicle based on monocular vision
CN109800689A (en) A kind of method for tracking target based on space-time characteristic fusion study
CN110232389A (en) A kind of stereoscopic vision air navigation aid based on green crop feature extraction invariance
CN108596165A (en) Road traffic marking detection method based on unmanned plane low latitude Aerial Images and system
CN102855500A (en) Haar and HoG characteristic based preceding car detection method
CN103927758B (en) Saliency detection method based on contrast ratio and minimum convex hull of angular point
CN107230203A (en) Casting defect recognition methods based on human eye vision attention mechanism
CN110119726A (en) A kind of vehicle brand multi-angle recognition methods based on YOLOv3 model
CN106297492B (en) A kind of Educational toy external member and the method using color and outline identification programming module
CN109145708A (en) A kind of people flow rate statistical method based on the fusion of RGB and D information
CN104636749A (en) Target object detection method and device
CN102930251B (en) Bidimensional collectibles data acquisition and the apparatus and method of examination
CN107729853A (en) A kind of automatic identifying method suitable for the narrow tuning drive gear formula instrument of transformer station
CN104318198A (en) Identification method and device suitable for substation robot patrolling
CN107392929A (en) A kind of intelligent target detection and dimension measurement method based on human vision model

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