机器视觉光学入门
第2部分

作为AIA认证的视觉专业基础课程的一部分，Greg Hollows, Edmund Optics的机器视觉总监，教授如何定义光学布局的基本参数，如何平衡系统的视场，分辨率工作距离和景深。

高级培训

订阅获取高级认证视力专业培训

点击这里

考试地点

找到一个认证的视力专业考试地点，测试您的视力培训知识。

点击这里

课程幻灯片

点击下面购买。

数字基本

数字先进

点击查看会议记录欢迎来到AIA的CVP计划的两个基本光学版......没有讨论我们只是完成的MCF是解决问题并进行了对比的Britt迈出了一步，现在开始进一步迁移到理解转移功能是有效的旧鞋子光学的适合在这个最终解决方案中，重组费用是您选择的传感器，它看出它的频率看一些有关光学组件的不同草和信息，以及超越动作我的良好选择您的申请与这个问题有什么更好的MCF我们有两种不同的MCFcurves here and on the vertical axis were actually the contrast your children are percent if you refer back to part one of this presentation you’ll notice as he went through there actually 20 to 30% was significant move says big deal in array all the way to 100 that the really beyond the possibility what she seemed in Houston meditated 0% frequency on this Pakistan here we have frequency laid out from zero hundred and this scenario to be going from basically infinitely large pixels down the pixels of about 5 µ in size the underlying pairs per millimeter pixels together make the line care that is in 10 µ 10 µ into one millimeters 100 times scanning is 100 line harasses uses a frequency range moving across the sensor in the mail here for your fifth you can make run pixels and Solon seeing Suzy pixels and have legal frequency and this gives us the ability to look these different versions run in the blue labor percent two different lenses and scenario and a would be going out there at this underlying pairs shall how we get different contrast solution levels dirt for back to part one we had an example were looking at those gel capsules very large objects very low-frequency objects AV rating on here somewhere simply sit what is the better lands insurance are you producing contrast in our example we would say the blue lands and being just a higher level mind for house and low-frequency requirement were looking large objects out of the right choice in the scenario here were getting a very very fine detail on a semi-play off of understanding frequency the sensors ever using and going that altogether sensors producing for us in life for CNN were spending that money on the sensor we want to push it all the way limit need all the detail you want to see out of it the bread will become the better choice because are moving into the space you to the blue line is no longer producing Hani contrast that details the red lines in the case of the right sure it’s in the middle are doing a variety things remain get push that it’s a new interpreting your application that each tells you and see in choosing right product is a simplified version of the curve for and keep expanding then again a more help here as we look at things understanding how it goes now something everyone understand here that we have to keep in mind is that the sensor and system also produces an end here everything your system produces an MCF cabling the computer that you use the lighting will contribute this an easy when it is the ability for anything to reproduce contrast in the system so if you lie highly unshielded cable in your system and you a lot of high power things running next to it you’re in no ways from those other things in your system and will reduce that signal information is going to signal and through the system Kikkoman this is the ability to reproduce the signal of what you’re seeing the sensor has this to visit an example from a Kodak sensor from nanometers provided by Kodak showing how their systematic is the limit of level sensor you using 9 µ pixels 55 line harasses the limit that’s sensors ability to refuse contrast to delete sees we go to the limiting marriage sensors ability to reproduce contrast you have the Lindsay the hundred percent go to zero as he moves to rehear we get different effects that are going on a different frequency levels now if we take it Lynn step further we had a lens is ability to reproduce contrast the sensors ability to reproduce contrast we – use system-level simulator reproduce contrast the well Reimers you discussed earlier if a system has 10% know anything at the sensor level and we a Lanza going is 10% contrast worsens you go to zero see that effect here all way up to hear the camera system to get of the lens system skinniest up 10% the camera of the sensor struggling to reproduce that basically Sierra Leone you are here since why the importance you imagine if we these objects to get this single level of much higher anything else in a degree in system is Kennedy Mueller depend on whole systems capability by a significant amount some very important to pull the pricing is greatly missing everything’s can have some effect on this issue roles system and it has pointed out as pixels get smaller and smaller these affects the greater Hassler those kind of bugaboos that the Internet you’re really have to pay attention to nature you don’t get caught with the just going to grill small pixels on on a low-cost sensors can adjust get you everything she want there are trade-offs there’s good reasons for those products to the out there but it’s understanding Wendy’s array product that) you see the desired results and system building facilities or mortgage for is this concept of MCF and understanding effectively light is are another example here asserts showing it helps you different lenses that are doing the same thing in different results at different frequencies as we saw on that random blue line graph to size ago or three size ago and surf I think the working distance color that are denizens system – number all will start affecting resolution so this example here we have to lenses lens that AMD and real look at the center in the image in the corner the image and relocated to different frequency levels will see some interesting things for recur here is very hard for anybody to see visually medicinal 5% contrast difference here between the two lenses lens be in the scenario on the right in the center has 5% higher contrast this particular resolution as we move out to the corner that system and we solstice on the lenses reanalyze earlier we get a minimum Roloff in that contrast the lens to be significantly lower about 35% lower than so we see this change from the center of the system one of the image seen things are those who start increasing the frequency greatly by three or we go back to the center of the system and we all sons see this sensor a lens be a strong are below the center of lands they were at the lower frequency lens be with higher research seeing some in the shifting and skeletal on precarious at the corner we ask you see a significant difference your with a philosophical white screen and this one as 50% higher contrast nonetheless this is approximately 10% contrast notice the ability to visualize this is probably low very very difficult see in this if you get down to the limits a 10% might seem like a lot to you and I will redo really this is 10% of something 10% contrast and imaging systems very difficult going to stop to 20% in the scenario is much much better get a lot more capability on the system now some things that understand here remember back on that on the first line to be had here the center of lands the recession the best in a lower frequency level Dallas had to do with the design the lenses designed is something very very specific know he actually looked of these lens be is actually more expensive lens on Wednesday the just finally caught Sir low-cost energy the soon I get what I pay for my team will get a better product windshield I can analyze easy to get the same focal length and similar specifications what you find is lens be with society something very very narrow was design work of one working distance exceptionally well what was pushed outside of his design. Reread quickly and that’s what we saw here this was an essay being used as general design the product could focus there could be US there but the design is very narrow area will be moving to its exact spot actio forms Wednesday when you do the same two locations percent to do with the resolution level camera that we are seeing the details the rehear but is also using an incorrect working distance from where was designed Soviets are seen these variations do work only three things and reason for showing this example is research stepping through NTF her analogy that you see an example of these resolutions in contrast look like this research talk about hers for skirts can be somewhat deceptive you locations they will maybe that is good enough for that’s not enough gaining some ground with contrast levels the playing in those differences between them is really really important Linnaeus look at her which is just a bunch of lines on some graph paper video: altogether another set of crazy examples here is the Senate Friday and everything is seen as an shift terms as we look at them is looking for the All-Stars hard to go altogether on a lot of different things we talk about resolution in contrast all long time the singing about the start target Uribe US Air Force target have different frequencies online the station cut smaller and smaller and closer and closer to the center of the rolling one discrete going in us the start target we get continuous frequency changes to go from the wider area VA error areas will family apply she would absolutely can see can continue his change and we can see in all directions ridges really interesting the 11th the and targets yet looked at Lake the Iraqi ruling inside the cover the entire field but it was one frequency cycad seachange both frequency and every direction across the given field area but if we take it one step further place the stars all over target we can see different resolutions and repositioning of the damage and we can see in all directions for anybody a half glasses and stoppers with astigmatism the lender stamina saying the union different resolutions and horizontal and vertical planes by the way are always do things optics in a different they will have those sort of promise real get different resolutions in different orientation and that’s why certainly a start target we can see all the orientations me the difference with one of the other sometimes you don’t. This elitist actually look at the MCF errors coming up in a few slides POC all the different frequencies at one time in all the different places on the image at one time in all the different orientation as you see some cases you might get a lands that works really well in every position except for one and that one physician maybe the one that’ll get shoe not let UBL to do the performance issue with one of the slightly lower at all places and that low when might be the better choice and I guess and the complicated portion of choosing the lens wasn’t your usual quick and see some live performances understeer them over to look at here again is to lenses for real look center as you look at these they are fairly close them unlocked is a little bit sharper we can see all the way down the middle one of them are looking bottom middle and we see it’s getting loaded blurrier on the right-hand side in the last number look out into the corner the image asked what’s going on and limited do is only in large what’s going on here the lens on the left is is better than the right significantly as we start looking at these spots will literally 10 look at here as we blow them up the larger is what’s going on we can see a lot of these effects all once and this ties together everything we’ve been talking about will does the left can image we can see fairly far down to the middle not as you could in the center spot is not revolving as well as one middle earlier but we can see fairly well down the middle and the best focus areas with issue look at the one on the right here even the low-frequency we can see a fair amount of blurring in all directions is a variety of issues that are going on here from color issues the something called home on a variety of other things or have in your treating the focusing so even at low frequencies this lens is struggling to maintain the resolution only when again there below entire attention is something else you can see a little bit here this oblong shape defocused areas if you really see here you can see there’s black or lines here in prayer lines in this direction black here astigmatism is occurring lands regain the focus basically Marland mansion another serene different resolutions one way and in the other that is directly related to the lenses we go. In these low-frequency years urlesque on in the 70 keep in mind and look for any systems on your focusing out the corn target like this rate for showing and authentically benchtop tool and should have just for benchmarking systems this will allow sauce understand where he is seen some of the slides that are coming out now there’s a lot of different things that can affect overseen and MCF curve right now are just talked with theoretical side of this lenses to compare one lens to another to see a there actually performing begin resolutions you want out of the camera system so the sinks are working distance we would also called on the optics for the conjugate distances that’s how far were actually focusing out to in focusing lens back on the sensor the wavelength range color can do an awful lot to your system in the different colors happened a lot of things were as yet f-number can help us Curtis depending on lens design is a variety of other operations redesigned when you control what is not on here is also the price-performance ratio this Feenstra and including the lens which will she bring the number of lens elements again put in the system and metal also constrain things like the type of metal that shoes were balancing the system machine vision systems are generally used to check tolerances on some other object others for scratches or signs assist things like that the see how close to the nominal design actually works vision lenses have no differences therefrom anything else it’s manufacturer you have some amount of tolerance stack up system that needs to be addressed in the more tolerance stack of problems that you have used in the more your image to Paris so things like that will be taken of tolerance is generally dictated enough controls on your manufactured and that will limit the ability for the or enhance or limit the ability lenses capability performed how much this tolerance are loosened worksite which will drive cost those literally things together the don’t see how performances effective so this is a true and the curve in a variety of things in here to look at will and will notice your looks somewhat similar to the the generic will earlier a couple different color line will show the more separation as you lenses have more separation in them but right now are using one that is pretty much a well-designed lens for this working distance we know level problems but there is a red green and blue settle lines here that are present the center center lands which is the blue .7 feel which is about half the full area 50.7 all of the object and that the green line in the blue line which is basically the corner lens estimate lenses see how over performing anyone in these positions overseeing the center based in the bottom middle in the corner performance across frequency with contrast very similar that start target you’re looking at visually see what’s going on this unbecoming graphical representation of what you will notice here as well as the black line on this one pretending to this one a little bit more loosely and address what’s called the diffraction limit this is as lovely thing will physics gives to us and says this is the limitation for contrasting get anywhere in the frequencies the things or involvement that there the wavelength of light that you use the f-number lens constant three data point echoing nothing more nothing less now what does that mean that the legal limit for the lens can do 50 come back and say I want a Lanza works at this frequency an island performance of here and you’re not willing to change Johnson’s the Constantino will change the f-number the wavelength if you’re using cannot adjustable outline some of the things that I can go any further enough from the lens design perspective how much money throw a Anthony keep in mind will be moved to small pixels is becomes very difficult Flagyl notice this black line to get a higher frequencies higher frequencies means smaller pixels that line continues to drive towards Europe going with you get that office to get allowances on the f-number small in the f-number higher that line goes but there’s other problems that come in new that will talk about later the example that we have here in the slides so on the car were using C 60 nm light this is rent laid only on the bottom it’s way late notice said in the red light scenario are moving much closer to the diffraction limit primarily visual visible to see the differentiation Erickson so close together and then your flight one a quick takeaways going monochromatic will generally give you better performance in the system.tends in the way it was assigned it’s a good rule of thumb green light is a great place to be whenever possible pieces any give you slightly better performance in the diffraction limit there is are wavelength is shorter diffraction limit gets better is our overall limitation or Lynn moves up may say well I can age you going sure shorter blue and purple when things like that is also a problem with the glasses being a control things in those low it’s hard for the design the function see usually a trade-off is a good place to be are trying to maximize everything so you see separations your bitching light red cell First Interstate ways monochromatic deadly better West Mason mother changed students is give you an understanding what goes on the some things with aberrational effect with color depending on maybe do a design we in a red green blue circle here right wing blue circle here is light is getting focus through the system on the I can show the soul with some coming off the some going to different wavelengths are focusing different positions away different distances out how well the design is made rolled to get all those in the same spot the same distance is eight is possible is really really important bulimia notice featuring these two circles here the greenest focus a lot later in the scenario but the blue is adding a lot of extra scanner in this case is blown out to much bigger size so overall light might have a nice tape spot in one color here in the overall spot amusingly light is very day reach one of these overhears a little bit bigger set of the blue everything inch screen lens and resolution that’s not to be confused with the fact that I could focus any one of these tighter if there were alone this is no way late scenario what you’re really trending or after is this the things that come into play here is the types of glass to use the more expensive or exotic mature like enjoyed bringing system in most cases the more I can control its was that you ask lost tolerance errors can do this to work I have very loose tolerances only design I made in this case to matter what I do because I’m getting too much shifting in my system some naval control things as well my spot get very day people - pixel units as big as this red circle make as much energy is coming on the blue circle here that’s an entry blurring in my system and that computer problem so controlling for wavelength and nine costs system the so we see with color fax nausea was one of think I’ll throw color this is an I move out from the center not only do I get this spreading are you focusing the site focuses way well I think the colors and scatters across the systems the unit manager via resolution contrast fairly good close to the passable physics limitation pixel decides the red one in the red green blue is also Celinda that same spot in the scatter across the media blurring this is little is a muscle now the next pixel problem my resolution contrast my ability to reproduce resolution contrast island romped because this is information is scattered from roles five pixel you can see why here few monochromatic this in limine this affects me things better. It’s only also goes on here and this goes back to that example the AEB lens where one was better in the middle and worse the corners and when the resolutions the way late scenario is done here at the bottom of the screen change a list better in the this is warmer changing working distance we same lands before both on as you look at the top of the screen we see some you change your what I wanted that refocus the lens suite distance it’s not happy at its well outside cesarean range this one here is working at 14 inches this when you’re working about six MEC that the working distance change on this is push the design so far outside what it was me do not hit all the performance of the lens Ronson you can see it being separation in the performance levels are hearing here blooming the middle still fairly good green is that means seven field is coming down to read the corners separate allotment owners you’re not that you green line the two red ones as we talk about his statement system and things like that direction all dependency on the Wednesday show in the system so that in the start target images that we looked at a little while ago that we can work better resolution one dimension in the other is exactly what you’re seeing is that fall off there is being separated by this directional dependencies is going on in this red line here is very pretty detrimental system and driving me down for what you’re trying to do this is the equivalent in many cases is taking opening spacers on the back of your lens the need to focus closer than it was designed in most cases not all that most cases of the lens has a certain amount of focus range to it because outside of that range is not designed to go there performance will generally suffer and suffer very quickly but also lost its combines and the things we just looked at together we now changed input that shorter working distance assigned your of the bottom systole very low performance away like we go back of here and we go back to the monochromatic situation we saw on the first lives in as all the performance came back to as the color aspect of it has been removed from the way late scenario Daniel LeBon will you other artifacts and another angel and we energy in their formants in the system the lending notice cell is a diffraction limit is lower than here so the absolute law physics limit is lower than the performances higher than the end of the day the physics limit is not much driving is to make your decision the actual performance of the lens seeing in a juggling act blending to be cautious of and it has been mean by on different lens specifications that are out there sometimes they specify on the absolute laws of physics limit in the Daschle formants the lens cautious of those things as you can see here that would make this lens the better balls physics limit is better than this one this is after the better performing will now take the big step back from that my recommendation we don’t shut choose a lens is designed to this working distance in this field of using or trying to achieve specifically cousin of performances Wally better since round lens designs at annual maximize which system let’s I’ve taken a step further now rented say this and go into debt at the same time right now up until now we’ve been talk of everything being flat plane now rests are looking at depth in the monthly notices interesting about this to start going through depth of field in our resolution thoughts is depth of field to get better and better depth of field you generally want to me the lens aperture of setting to a higher number of going from 8 to 1111 next 16 to extend the depth of field were refined with resolution remember I said the three think the dry theoretical limitations of the wavelength of the light constant in the f-number when you do this calculation the smaller are the lower the f-number better the resolution so you get better resolution theoretically you wanting a from FA 5.6 F4 get better the field you go FA 1116 give finding your opposite direction and this is my least point to see if they mention and message you open and close the are or shall see the center of an opening closing bigger and smaller in size what you’re effectively changing is the diameter of the lens itself the focal length is fundamentally staying the same alleged to using zoom lens or something like that but in any fixed position the focal length of because in the area diameter with the aperture of the lens and that allows us to do ulcers of fun things allows us to control light measuring a few slides your depending on the lens out through resolution and also sure you how it can actually hurt resolution depending the land and effect depth of field understand the number of the lens has a lot of effects that could be unintended if you’re not prepared to deal with with angle on there keeping all that in mind over here look to see how it affects everything so to go with the smallest you will sponsor as exhibit a constant wavelength away the number but the real way to look at this is this example here so in our aunt Vicki can remember this from and in section 1 literally can’t eschew Lot Jackson coming closer and further part as an object is image to the lens if we had a wide-open theoretically beginning Chivas small I saw it and that would be represented by this curve here as we start closing down that Iris get more traction in the city system and easily has a wider profile bottom, it’s downloaded we just getting Daisy Baker blurrier stop remember back before we talked about this two spots coming together Nika bleed over register closing on your number eventually all those spots will bleed over together comparable those pixels and you should resolution that your real problem there’s lots of things that can come in your set always at limiting factor is the fracture on his mentioned earlier is not the only limiting factors lots of things that go into it that have to do the pillars in the design and other things with manufacturing tolerances and cost structures and written adult normality everything we need to be a bulldozer resolutions that we want something to consider that there biggest light years in you might have to lenses that have the same number on but they might not and now the highest level performance they can achieve can be the same it doesn’t mean at the end of the day because of the things were talking about specially when it comes than the designed manufacturing is are actually going to get the performance levels that are there and that’s really Laurent enhancements are looking at. But an example here it seasonings a very simplified product and will look to see how it affects things bargaining build a look at the modulation transfer function as so many her video lenses in the sea on a little security camera or maybe like a alleged can personally that he’s a little small guys even carried over some of the higher in hot camera phones have things that are similar to this on them and to their societies in the ostrich that we get a lot of things have to be removed whether it elements of weather tolerancing and all that and you don’t get as many degrees of freedom in the design to the other do what you wanted you a feeling work with few lens elements inside as opposed to many no limited by the number of glasses because of cost constraints to make it performance it looks like this again with the MTF curve with a frequency and interests here you see his black client the top is fairly five for this wide-open f-number best to the calculator will numbers really really good we get real high levels of performance to the actual performances fairly well one of the nice things as we start closing down the picture we see the performance level, and the reason for that is when you have a wide-open have to charnel lens you’re using very oblique rays it’s very hard then them all back to you want until effectively Khmer back to the drawling of it on the board in session while I’m we is off axis rays are hard to control so you can effectively trying them out they talked about earlier to get rid of them in fact create a uniform light level crossings higher sensor in this scenario here is holding on in are doing the same thing were taken is oblique rays are hard to control or eliminate getting them done things control system and thus you don’t get all these extra facts and her performance level start rising mistake as a step further so safer second now okay I’m not you can stream by being a small package size is development some cost this using very low cost lenses as an extreme example you son and I expect to be a little bit more industrial is to have more capabilities to it I Nasional you design something as I have more flexibility more degrees of freedom they can come much for fractional and let’s say for you go back and look at the last line is a reference of for being here see much lower performances constraints I eliminated is constraints advanced my design on very close the diffraction limit now thoroughly very interesting occurs like much better forming lens after the war they search shutting down my Iris at a F 12 to 16 note is that the overall performance search getting right up against diffraction limit as it pushes down my performance the same thing happens here as I move through the diffraction limit comes down my role performance and sentiments very interesting years you can start seeing were prior and 60% contrast the limit heretofore have 16 we barely getting zero an error dissolute about 01 two thirds the performance the things of shifted greatly at this point and that is hurting my role formats of my system remember ergo after some calculator level than filling with your sensor with the frequency level this can be problematic by closing the lens now and that related depth field things you feel you want to as a mansion you want to shut the lens down to get better depth for can see why in a moment something that we should really understand about depth field it is then the year-earlier specifying with a resolution in mind became saying he could depth of field you have to specify some level of detail the one to see I need to see objects of this size or I need to see a scratcher effect at this size at this level way because good depth of field is all relative if yours trying to discern is something black-and-white nervy objects it’s one bit member the till example from earlier your trainer be a 2-D data matrix is on till they can be at different heights that the wildly different requirement from resolution point you and easy it seems you have to start specifying things like that to go full Celeste limited visual version of how depth of field works in understand to lenses here and these red lines in by the same physician this is the acceptable amount of blur in my system the acceptable amount of defocused from my spine thinking see in my system and what they do is read at this position here on both of them wears best focus my pictures open very wide we had these.a lines that come to the best focus and really crossed to the acceptable level that’s light depth field that’s for things that are blurry and either and would still be out of your resolve beyond not there to glory for me to see on either side so I’m willing allow myself to a very blurry objects angle gets very day I went not very blurry objects they can small as you start closing out here picture in a system of lines that Anglicans very narrow with a pass your best focus Musial much extended range of the few liking in this but this gets down to see how fast is the golden of focusing on changing as I go through my system and that it’s only how much distance of the Maxi only get good focusing my system in the one of those things is all about these homes that are created and how lowlight even if it’s outside: information up here is now being focused through that spot some getting all this detail from near to hear that see before is getting focused at this position that can’t pull it out effectively they close it down theoretically better so we see the world we are looking at MTF curve is a certain clothing now my picture I resolutions going down theoretically me depth field going out for the run counter to each other active the problem are looking at things but so will look at some examples here of how we actually test depth of field in the system we use target in this case and targets are was a good thing the happiness that 45° wedge and got lines on C beings is going through focus in the system and some of them are in this case running in this direction summer running long-lived on the terms well see this in a little bit better things on this data target is there’s a scale of lenses matter of the actual distance ever seen only years ago from blurry clear back to blurry world your lines

机器视觉光学入门第2部分

机器视觉光学入门
第2部分