CN106097308A - A kind of New energy electric vehicle based on machine vision charging hole detection and localization method - Google Patents

Abstract

The invention discloses a kind of New energy electric vehicle based on machine vision charging hole detection and localization method, obtain cradle image first with vision sensor；Then the noise jamming introduced to picture signal for strong electromagnetic charging system, the median filtering method that have employed classics removes noise；Owing to cradle image has the features such as background complexity, brightness disproportionation, reflective, target characteristic is few, conventional fixing or adaptive thresholding method is difficult to effectively split ideal object, have studied two-stage image partition method based on HSI color model again, i.e. obtained the area-of-interest of rough grade by the thresholding of the conversion of HSI color model and Hue component, obtain high-precision target area by morphological operation and Canny operator edge detection；The final feature extracting target charging hole.

Description

A kind of New energy electric vehicle based on machine vision charging hole detection and localization method

Technical field

The invention belongs to field of image recognition, particularly to a kind of New energy electric vehicle based on machine vision charging hole Detection and localization method.

Background technology

Auto industry concerns lifelines of the national economy, drives huge vertical industry chain, is that China's Major Strategic props up Post industry.New-energy automobile, as modern automobile industry one spotlight, is the important engine promoting the sustainable development of socio-economy. " energy-conservation and new-energy automobile " as giving priority to field, is specified industrial development direction by " made in China 2025 ".By In the end of the year 2015, China's new-energy automobile recoverable amount reaches 58.32 ten thousand, increases nearly 170%, in explosion type development trend.

Different types of new energy vehicle, such as electric vehicle (EV), extended-range electric vehicle (EREV) and hybrid power Electric vehicle (HEV), it is equipped with the energy storage system needing periodically charging.Generally, by by energy storage system even Being connected to power supply, such as AC supply line, can be its charging.Although storing for vehicle energy before or after each vehicle uses It is favourable that system recharges, and supply line is manually inserted vehicle by current system requirements vehicle driver.Such fill Electrically there is many unfavorable factors, the most artificial hand-held charging gun causes poor efficiency and/or misses optimal charge, vile weather Under the conditions of vehicle driver is not easy to field operation, supply line is likely to occur leak current fault and brings potential safety hazard.

Accordingly, with respect to traditional artificial charging, vehicle driver saves manpower, time-consuming, peace in the urgent need to one Full automatization's charging modes reliably, to realize this link unmanned, rapidly and efficiently.But electric vehicle automatization to be realized Charging, how the key of its technology and difficult point are to the detection of electric vehicle charging hole and location.

Summary of the invention

The invention provides a kind of New energy electric vehicle based on machine vision charging hole detection and localization method, its mesh Be, by find cradle characteristics of image, to electric vehicle charging hole detection with location thus realize electric vehicle from Dynamicization is charged.

A kind of New energy electric vehicle based on machine vision charging hole detection and localization method, first, it is thus achieved that cradle Image, and be filtered cradle image processing；Secondly, utilize the two-stage image partition method of HSI color model to filter Cradle image after ripple processes carries out dividing processing, it is thus achieved that initial area-of-interest；Then, initial area-of-interest is used Morphological operation and Canny operator edge detection obtain high-precision target area；Finally, from target area, extract target to fill The feature in electricity hole.

Tone (Hue) specificity analysis of charging bore region: due to metal material at the core of hole, the most certain reflecting effect, Make this region be positioned at a range of same tone on the original image, be viewed as from human eye angle the most golden yellow or inclined Bright/or partially dark, belong to warm tones；And core edge surface position, hole, it is considered that it is celadon plastic material, belongs to dark cool tone, make Obtain this image and be significantly different from tone at the core of hole.The above, the gray value of the Kong Xin that i.e. charges in image Hue component with The grey value profile of peripheral region is at two different neighborhoods.Utilize this characteristic can realize charge hole and background in Hue component Separation.

Threshold value in the two-stage image partition method of described HIS color model determines that mode is to use maximum between-cluster variance Automatically take threshold value or take threshold value by calculating grey level histogram crest paddy.

Described employing maximum between-cluster variance automatically takes threshold value and refers to so that two class population variancesTake threshold value T of maximum:

${\mathrm{\σ}}_T^2={\mathrm{\ω}}_a{({\mathrm{\μ}}_a-\mathrm{\μ})}^2+{\mathrm{\ω}}_b{({\mathrm{\μ}}_b-\mathrm{\μ})}^2$

Wherein, threshold value T in the two-stage image partition method of HIS color model is by the cradle image after Filtering Processing It is divided into charging hole and background two class part, σ²For maximum variance, ω between charging bore portion and background parts two class_aFor Filtering Processing After cradle image in pixel belong to charging bore portion probability, μ_aFor the pixel average gray of the bore portion that charges, ω_bFor filter In cradle image after ripple process, pixel belongs to the probability of background parts, μ_bFor the pixel average gray of background parts, μ is filter Pixel population mean gray scale in cradle image after ripple process.

Described by calculate grey level histogram crest paddy take specifically comprising the following steps that of threshold value

A) the absolute grayscale rectangular histogram in cradle image range after statistical filtering processes；

B) valley between first and second peak values in absolute grayscale rectangular histogram, and first and second peak values is found；

C) global threshold split in use the valley of bimodal as threshold value, such as following formula:

$p(x,y)=\left\{\begin{array}{cc}255& q(x,y)\&GreaterEqual;T\\ 0& q(x,y)<T\end{array}\right.$

Wherein, (x, y) represents image after segmentation to p, and (x, y) represents the cradle image after Filtering Processing to q, and T is HIS color Threshold value in the two-stage image partition method of model.

Image after segmentation is carried out connected domain search operation, searches out wherein that circularity and area meet circularity and area sets Connected domain Blob of fixed condition, and connected domain Blob is carried out holes filling, obtain non-NULL connected domain Blob_fillup；

Generally circularity span is 0.8-1, and area is depending on practical situation；

$Blob\_fillup(x,y)=\left\{\begin{array}{c}255,\&ForAll;(x,y)\&NotElement;Blob\\ Blob(x,y)\end{array}\right.$

Then, by convex closure, non-NULL connected domain being carried out shape conversion, matching profile also merges connected domain, first fixed by obtain The charging hole area-of-interest of position is as initial area-of-interest.

Described as follows to initial area-of-interest employing morphological operation and Canny operator edge detection process:

Step 4.1: expand initial area-of-interest respectively and etching operation, it is thus achieved that two border circular areas, by two It is poor that individual border circular areas subtracts each other, it is thus achieved that annulus area-of-interest RegionDiff；

Step 4.2: use Canny operator that annulus area-of-interest is carried out rim detection, extract edge Edges, and Segmenting edge, it is thus achieved that line and circle；

Step 4.3: the circle using Tukey approach method to obtain step 4.2 carries out robustness matching, obtains target area Sub-pixel precision profile, be charging hole, high-precision target area profile.

Before the circle using Tukey approach method to obtain step 4.2 carries out robustness matching, first step 4.2 is obtained Circle judges according to the attribute of circle, rejects not rounded.

During described employing Tukey approach method, during matching, use the MaxNumPoints=-1 that at most counts of profile；Profile Close threshold value MaxClosureDist=0；Match point number ClippingEndPoints=0；Iterations Iterations= 3；The shear factor ClippingFactor=2 of outlier.

The circle matching of Tukey method robustness, idiographic flow is: weighting function is all one circle of 1 matching first for the first time, so The each profile point of rear calculating obtains weighting function, and the weight calculated as next time to the distance on circle, then by same step weight Multiple iteration, finally gives preferably circle.

Each charging centre coordinate in hole and area features is extracted from target area:

(1) each charging center, hole O (x in target area is asked for according to the following formula₀,y₀), the center of circle in hole of respectively charging；

$O_k(x,y)=(\underset{i=1}{\overset{M_k}{\&Sigma;}}{x}_i/M_k,\underset{j=1}{\overset{M_k}{\&Sigma;}}{y}_j/M_k),\&ForAll;(x_i,y_j)\&Element;S_k$

Wherein, k is charging hole numbering, O_kRepresent the kth charging center of circle, hole, S_kRepresent kth charging bore region, M_kFor kth Pixel sum in individual charging bore region；

(2) target area feature is calculated:

Utilize green theorem to calculate interest region area, then discretization obtains number of pixels in target area.

Medium filtering is used when being filtered cradle image processing.

Beneficial effect

Compared with prior art, advantages of the present invention be embodied in following some:

1) present invention employs the determination methods of two-stage detection area-of-interest, i.e. determined by rough grade, high accuracy Charging bore region, effectively location charging hole site.Total algorithm is simple, efficient, and the present invention uses HALCON software to enter algorithm Row validation test, the operation time is about 1104.61ms；

2) compared with traditional image partition method, the threshold value based on HSI color model of two kinds of present invention offer is divided Segmentation method more advantage.Wherein, for the image of shooting at close range, both of which is suitable for, and has good segmentation effect；But For the image of wide-long shot, if under conditions of light illuminates, maximum between-cluster variance automatically takes threshold method and reaches reluctantly point Cut effect, but under conditions of illumination is dark, a large amount of burrs occurs in image after treatment, and segmentation misses by a mile.Therefore calculating gray scale Rectangular histogram Wave crest and wave trough takes threshold method and is applicable to extensive condition (shooting distance, illumination brightness), and maximum between-cluster variance takes threshold automatically Value method is relatively applicable to shooting at close range and/or the brighter condition of illumination；

3) compared with existing manual method, this detection algorithm has the advantages such as accuracy of detection is high, speed is fast, reproducible, Meet being automatically brought into operation of robot manipulating task, the demand such as the quickest, reliable and stable well.And artificial charging by hand cannot essence Determine charging hole site, position, easily cause and miss the situation of charging and/or vehicle performance therewith is degenerated.

Accompanying drawing explanation

Fig. 1 be the present invention charge hole detection with location image processing flow figure；

Fig. 2 is image filtering schematic diagram, and wherein (a) is charging hole material picture；B () is the noise disturbed by salt-pepper noise Image；C () is the denoising image after 3*3 template medium filtering；

Fig. 3 is that rough grade determines area-of-interest schematic diagram, and wherein (a) is the region of interest ROI created；(b) be (a) The Hue component of figure HSI color space；C () uses maximum between-cluster variance automatically to take the bianry image of threshold method for (b) figure；(d) be B () figure uses calculating grey level histogram Wave crest and wave trough to take the bianry image of threshold method；E () is the interested of extraction from connected domain Region through shape conversion, rough matching；

Fig. 4 determines area-of-interest schematic diagram for high accuracy, and wherein (a) is that the expanded corrosion of Fig. 3 (e) figure remakes after the recovery Annular area-of-interest；B () is (a) region under gray level image；C () is Canny rim detection after, through the Asia of robustness matching Pixel precision edge.

Detailed description of the invention

Below in conjunction with drawings and Examples, the present invention is described further.

As it is shown in figure 1, a kind of New energy electric vehicle based on machine vision charging hole detection and localization method, including with Lower step:

Step 1: collected by camera, to charging hole original image, is denoted as Image0；

Step 2: Image0 carries out the medium filtering of 3*3 template, obtains the hole denoising image that charges, is denoted as ImageNoise, as shown in Figure 2；

Step 3: denoising image ImageNoise is carried out in units of pixel preliminary images segmentation, rough grade determines that sense is emerging Interest region；

Step 3.1: create region of interest ROI, limits image procossing scope, is denoted as ImageROI；

Step 3.2: transition diagram is as ImageROI to HSI color space, it is thus achieved that Hue component, is denoted as ImageHue；

Step 3.3: select appropriate threshold ImageHue to carry out Threshold segmentation based on HSI color model, by Hue component Binarization operation, makes charging hole split from background image, obtains bianry image ImageBin.Wherein threshold value selects there are two kinds Method: maximum between-cluster variance takes threshold method automatically, calculating rectangular histogram peak frequency gray value takes threshold method；

$\mathrm{Im}ageBin(x,y)=\left\{\begin{array}{cc}255,& G(x,y)>T_1,G(x,y)\&Element;\mathrm{Im}ageHue\\ 0,& else\end{array}\right.$

Wherein, G (x, y) be in ImageHue (x, y) position pixel gray value, T1 is can be by charging hole and the back of the body The separate appropriate threshold of scape；

Step 3.4: bianry image ImageBin carries out connected domain search operation, searches out wherein circularity and area and meets Connected domain Blob (usual circularity takes 0.8～1, and area is depending on practical situation) of certain condition, and carry out holes filling, obtain Non-NULL connected domain Blob_fillup；

$Blob\_fillup(x,y)=\left\{\begin{array}{c}255,\&ForAll;(x,y)\&NotElement;Blob\\ Blob(x,y)\end{array}\right.$

Step 3.5: carry out shape conversion by convex closure, matching profile also merges connected domain, obtains the charging hole sense just positioned Interest region, is designated as Image1, as shown in Figure 3；

So far, under the first stage, the charging hole of background image is the most divided out, but real image exist speck, Bai Hen, Textures etc. affect, and simple binarization operation cannot meet the hi-Fix of charging hole site, therefore for ensureing the centre of location Accuracy, by second stage process again with sub-pixel precision extract target.

Step 4: on the basis of step 3, uses the side that sub-pixel edge based on morphology and Canny operator detects Method, further hi-Fix charging hole target；

Two border circular areas are subtracted each other work by step 4.1: expand Image1 respectively and corrode, it is thus achieved that two border circular areas Difference, it is thus achieved that annulus area-of-interest RegionDiff；

Step 4.2: use Canny rim detection, extract edge Edges, and segmenting edge: line and circle；

Step 4.3: select edge, determine whether fitting circle according to characteristic, is not that the edge of 1 is rejected, wherein by Attrib Attrib=-1 represents that line segment, Attrib=0 represent oval, and Attrib=1 represents round；

Step 4.4:Tukey method is approached, and robustness matching circle after rejecting obtains sub-pixel precision profile, is The charging hole profile that high accuracy determines, is denoted as Image2.Wherein the parameter of Tukey method fitting circle is set as: MaxNumPoints =-1, MaxClosureDist=0, ClippingEndPoints=0, Iterations=3, ClippingFactor=2.

So far, charging hole target has realized hi-Fix, as shown in Figure 4, this target area of subsequent process Main Analysis Style characteristic and position characteristic.

Step 5: to every section of profile numbering, extract corresponding parameter (area Area, center abscissa Row, the vertical seat in center Mark Column), it is saved in array, completes the location of cradle；

Step 5.1: ask for each charging center, hole O (x in target area according to the following formula₀,y₀), the center of circle in hole of respectively charging；

$O_k(x,y)=(\underset{i=1}{\overset{M_k}{\&Sigma;}}{x}_i/M_k,\underset{j=1}{\overset{M_k}{\&Sigma;}}{y}_j/M_k),\&ForAll;(x_i,y_j)\&Element;S_k$

Wherein, k is charging hole numbering, O_kRepresent the kth charging center of circle, hole, S_kRepresent kth charging bore region, M_kFor kth Pixel sum in individual charging bore region；

Step 5.2: calculate target area feature, mainly utilize green theorem zoning area, then discretization obtains, real What matter represented is exactly number of pixels in target area.

Finally should be noted that: above example is only in order to illustrate that technical scheme is not intended to limit, to the greatest extent The present invention has been described in detail by pipe with reference to above-described embodiment, and those of ordinary skill in the field are it is understood that still The detailed description of the invention of the present invention can be modified or equivalent, and any without departing from spirit and scope of the invention Amendment or equivalent, it all should be contained in the middle of scope of the presently claimed invention.

Claims (10)

1. New energy electric vehicle based on a machine vision charging hole detection and localization method, it is characterised in that first, obtain Obtain cradle image, and be filtered cradle image processing；Secondly, the two-stage image of HSI color model is utilized to split Method carries out dividing processing to the cradle image after Filtering Processing, it is thus achieved that initial area-of-interest；Then, to interested Region uses morphological operation and Canny operator edge detection to obtain high-precision target area；Finally, carry from target area Take the feature in target charging hole.

Method the most according to claim 1, it is characterised in that the two-stage image partition method of described HIS color model In threshold value determine mode be use maximum between-cluster variance automatically take threshold value or by calculate grey level histogram crest paddy take threshold value.

Method the most according to claim 2, it is characterised in that it is to instigate that described employing maximum between-cluster variance takes threshold value automatically Obtain two class population variancesTake threshold value T of maximum:

${\mathrm{\&sigma;}}_T^2={\mathrm{\&omega;}}_a{({\mathrm{\&mu;}}_a-\mathrm{\&mu;})}^2+{\mathrm{\&omega;}}_b{({\mathrm{\&mu;}}_b-\mathrm{\&mu;})}^2$

Wherein, the cradle image after Filtering Processing is divided into by threshold value T in the two-stage image partition method of HIS color model Charging hole and background two class part, σ²For maximum variance, ω between charging bore portion and background parts two class_aAfter Filtering Processing In cradle image, pixel belongs to the probability of charging bore portion, μ_aFor the pixel average gray of the bore portion that charges, ω_bAt filtering In cradle image after reason, pixel belongs to the probability of background parts, μ_bFor the pixel average gray of background parts, μ is at filtering Pixel population mean gray scale in cradle image after reason.

Method the most according to claim 2, it is characterised in that described by calculate grey level histogram crest paddy take threshold value Specifically comprise the following steps that

A) the absolute grayscale rectangular histogram in cradle image range after statistical filtering processes；

B) valley between first and second peak values in absolute grayscale rectangular histogram, and first and second peak values is found；

C) global threshold split in use the valley of bimodal as threshold value, such as following formula:

$p(x,y)=\left\{\begin{array}{cc}255& q(x,y)\&GreaterEqual;T\\ 0& q(x,y)<T\end{array}\right.$

Wherein, (x, y) represents image after segmentation to p, and (x, y) represents the cradle image after Filtering Processing to q, and T is HIS color model Two-stage image partition method in threshold value.

5. according to the method described in any one of claim 1-4, it is characterised in that the image after segmentation is carried out connected domain search Operation, searches out connected domain Blob that wherein circularity and area meet circularity and area imposes a condition, and enters connected domain Blob Row holes filling, obtains non-NULL connected domain Blob_fillup；

$Blob\_fillup(x,y)=\left\{\begin{array}{c}255,\&ForAll;(x,y)\&NotElement;Blob\\ Blob(x,y)\end{array}\right.$

Then, by convex closure, non-NULL connected domain being carried out shape conversion, matching profile also merges connected domain, by the first location that obtains Charging hole area-of-interest is as initial area-of-interest.

Method the most according to claim 5, it is characterised in that described to initial area-of-interest use morphological operation and Canny operator edge detection process is as follows:

Step 4.1: expand initial area-of-interest respectively and etching operation, it is thus achieved that two border circular areas, by two circles It is poor that shape region is subtracted each other, it is thus achieved that annulus area-of-interest RegionDiff；

Step 4.2: use Canny operator that annulus area-of-interest is carried out rim detection, extract edge Edges, and split Edge, it is thus achieved that line and circle；

Step 4.3: the circle using Tukey approach method to obtain step 4.2 carries out robustness matching, obtains the Asia of target area Pixel precision profile, is charging hole, high-precision target area profile.

Method the most according to claim 6, it is characterised in that the circle using Tukey approach method to obtain step 4.2 enters Before row robustness matching, first step 4.2 acquisition is justified and judge according to the attribute of circle, reject not rounded.

8. according to the method described in claim 6 or 7, it is characterised in that during described employing Tukey approach method, adopt during matching With the MaxNumPoints=-1 that at most counts of profile；Profile closes threshold value MaxClosureDist=0；Match point number ClippingEndPoints=0；Iterations Iterations=3；The shear factor ClippingFactor=2 of outlier.

Method the most according to claim 8, it is characterised in that extract the centre coordinate in each charging hole from target area And area features:

(1) each charging center, hole O (x in target area is asked for according to the following formula₀,y₀), the center of circle in hole of respectively charging；

$O_k(x,y)=(\underset{i=1}{\overset{M_k}{\&Sigma;}}{x}_i/M_k,\underset{j=1}{\overset{M_k}{\&Sigma;}}{y}_j/M_k),\&ForAll;(x_i,y_j)\&Element;S_k$

Wherein, k is charging hole numbering, O_kRepresent the kth charging center of circle, hole, S_kRepresent kth charging bore region, M_kFill for kth Pixel sum in electricity bore region；

(2) target area feature is calculated:

Utilize green theorem to calculate interest region area, then discretization obtains number of pixels in target area.

Method the most according to claim 9, it is characterised in that use intermediate value when being filtered cradle image processing Filtering.