Jason Barnes asks leading machine vision industry figures what they consider to be the educational barriers to the technology’s increased uptake by the ITS sector. The recent rush by some organisations within the ITS sector to associate themselves with the term ‘machine vision’ underlines just how important the technology has become in a relatively short space of time. However, despite the technology having been applied in certain traffic management applications for some years, there remains a significant s
Marc Damhaut, CEO of Euresys
Jason Barnes asks leading machine vision industry figures what they consider to be the educational barriers to the technology’s increased uptake by the ITS sector.
The recent rush by some organisations within the ITS sector to associate themselves with the term ‘machine vision’ underlines just how important the technology has become in a relatively short space of time. However, despite the technology having been applied in certain traffic management applications for some years, there remains a significant shortfall of knowledge as to what machine vision is and what it can do.
While this is inevitable wherever new technology is applied, in some respects historical attitudes within the ITS sector are also responsible for the situation. Traffic management shares much with the security sector (for instance) in that it has become used to cameras being commodity items and a constant, downward pressure on price for a given level of performance. This drives a cost-consciousness which is absent from more traditional machine vision markets such as production processing, where it is more readily accepted that solutions have to be bought up to a quality and not down to a price.
Machine vision suppliers have moved to adopt the technology and address market needs - as ITS International previous detailed – although other, related issues, still remain. One is the presumption on the part of decision-makers of a camera’s capabilities based on historical norms – it is rather too easy to consider machine vision as ‘just’ an enhanced or next generation of camera, rather than allowing new ways of doing things. Another is that decision-makers can too easily decide that certain features are ‘must haves’, which then places undue restrictions on suppliers which can prevent them from engineering the most effective solutions. Yet another is caused by the complex nature of machine vision itself: specialisation at both the camera and component levels means that many of the machine vision sector’s experts actually have little opportunity to communicate or collaborate with end users.
As with so many ITS-related issues, the answer is education. The presumption is that the machine vision sector has to teach the traffic management sector about what the technology can do - but the reality is that education needs to become more of a two-way street. Traffic managers need to give some machine vision developers and suppliers a greater insight into what they do. The question arises as to just what might be the best way(s) in which to improve general levels of knowledge. As ever, a survey of the machine vision sector throws up as much agreement as discord.
7084 Euresys. His company, which supplies image processing libraries and vision cards, divides its interests into two broad areas of application: machine vision (for traditional manufacturing processing) and video surveillance.
“My instinct would be to place ITS within the video surveillance segment but the reality is that for some applications ITS may in fact sit somewhere between the two,” he states.
Finding the right forums for meaningful interaction is difficult, he feels: “We’re a small company and we can’t hope to educate the whole world on, for example, our HDcctv/HD-SDI products.
Damhaut was surprised at seeing ITS customers at last year’s6963 Vision Show in Stuttgart but notes that there was an ITS section in the ITE Show in Yokohama – Asia’s largest machine vision exhibition – a couple of years ago.
“It may just be that Japan and other Far Eastern countries are more advanced in the application of the technology for traffic and transport applications. In the past we’ve seen showcases for the medical sector at VISION so it’s not beyond credibility to see something like that happening in the future for ITS.
“A better forum for machine vision/ITS would be great and is, probably, needed. Perhaps in the interim, social media, such as the various discussion groups on sites such as LinkedIn, can provide an infill.
6965 Gardasoft Vision. His company has recently produced a ‘white paper’ designed to address some of the wider misperceptions about lighting systems within ITS machine vision applications.
“Too often we see lighting systems being specified at the eleventh hour, when really they should be considered far earlier in a project. There are functional and commercial advantages to better and earlier integration; not only do you end up with a system which may cost appreciably less but you can also achieve better performance by being able to work out the interplays between cameras and lighting. We certainly need a greater awareness of capability versus expectation.”
Downey doesn’t see it as a case of laying blame at the traffic sector’s door. An element of teaching and learning is always to be expected as technology vendors release functionally superior products, he notes.
“With some of our high-tech products, as with other technology vendors, there’s an education process even within the expert community. Vendors wouldn’t be doing their jobs if they weren’t releasing products which push the boundaries – otherwise we’d just have a ‘me too’ environment.
“The point is that the differences in knowledge are always going to be more apparent when technology crosses sectors, such as from machine vision into traffic. But vendors also need to understand different vertical markets and have to ensure that benefits are appreciated in respective sectors. Forums are great in that respect, in that customers bring their needs and we bring our products and solutions. The debate is whether the forums we have at present are adequate. Whatever the forum style there should be onus on technology vendors to ensure that they explain and illustrate their products within an application/solution context.
518 Allied Vision Technologies.
“‘Traditional’ security camera manufacturers are less able – and less willing – to customise unless large numbers of cameras are being ordered as they’re large box-shifters selling commodity products. The ability of machine vision camera manufacturers, who’re usually smaller and more nimble companies, is much greater.
“But there remain areas where the machine vision sector needs a better understanding of the transport sector’s requirements, such as changing light conditions and temperature ranges. Machine vision system manufacturers have been targeting this sector for a while now but I think they could still do with a much better understanding of what happens behind the camera – how the information it provides is used. That could allow us to come at a problem from a different direction entirely.
1679 Gatso to develop bespoke products. Generally though, says COO Lou Hermans, this doesn’t happen.
“My experience is that people tend to buy what’s available and the situation with Gatso was exceptional. A complicating factor is that red light and speed enforcement legislation varies widely from country to country and has to be taken into account. Another factor is dynamic range, given that light conditions in traffic applications can vary so widely. It’s for the system integrators to lead any education effort, I think.”
7179 Pleora Technologies divides the ITS sector’s use of machine vision into two broad categories: traditional video-type applications such as traffic flow monitoring; and higher-end applications such as in-vehicle and licence plate recognition for tolling and enforcement purposes.
Traffic monitoring applications, he notes, tend to use commodity products and there are a lot of suppliers servicing a large number of vertical markets. The higher-end ITS applications tend to be more select as machine vision cameras can be triggered (an image capture feature which comes from manufacturing processing). They also offer very high frame rates - typically 100-200 frames per second (fps) versus CCTV’s 20-30fps - which improves the chances of a clear capture, and use uncompressed video. The last point is especially important because what might work for the human eye may be not adequate for automated algorithms.
“The electronic toll collection and licence plate markets are definitely getting the use of machine vision correct but there’s room for improvement in general traffic monitoring,” he says. “Traditional video is largely designed for human consumption but soon we’ll reach a point where we need more automated monitoring and automatic alarm generation.
The ‘Holy Grail’ is something like an Amber Alert situation where there’s been a child abducted and you’re trying to find a given vehicle which you know is out on the network right now. Throwing large numbers of people at the problem doesn’t work; using lots of people costs and humans are fallible. With machine vision the situation is rather different.
“But the machine vision industry could do better at introducing itself. There’s a perception that it’s too expensive and not economically feasible for some applications. That Amber Alert situation I described shows that cost is relative and we’re already seeing moves which will significantly shift the cost balance.
“There’s a big shift away from traditional computing architectures where processing goes on at the PC level. Embedded systems are starting to emerge which offer high processing capabilities in very small form factors which don’t need, for instance, the cooling fans and other peripherals required by PCs. That means they can be located in a roadside cabinet or up a gantry or mounted with a camera. That’s a big upward shift in intelligence at any given point on a network and big downward shift in cost - and it’s happening now. We’re already seeing these solutions in the military market and I think that inside two to three years we can expect something of an explosion in the civil sector.
“But how do we communicate all this? Chat groups, I’ve found, are a good way to get information out but they’re much better if people already know what they’re looking for. Perhaps a better way is to present at more industry-centric shows, such as those for ITS; and given the huge potential applications of machine vision, I don’t think it’s unreasonable that the technology gets its own, dedicated conference streams.”
822 Stemmer Imaging’s Steve Hearn says, by comparison with three to five years ago, system integrators are doing a much better job of defining end-users’ needs.
“The requirements we see are much more fully formed. The problem over the last few years has been that there’s simply been no money to invest in new technologies or projects. However the last six months have seen some positive movement in this regard.
593 Teledyne Dalsa’s product marketing manager Manny Romero says that a little education is a dangerous thing. He believes that pressure is applied from senior figures within government organisations who think they know what they need and then dictate how a solution should be arrived at rather than letting the experts do what they know best.
“It’s tempting to interfere when you have a little knowledge. An example is where the desired next level of magnification is stated to be ‘X’ but there’s no realisation that increasing resolution may not always give a better result,” he says. “There’s a good grasp at the engineering level but not always at the policy or decision-making levels.
“Right now the main constraint in the ITS sector is price, not system efficiency -it should be the other way around. Price point is based on what people expect to get from a feature point of view but management may reconsider targeted price points if they’re shown the benefit of advanced feature sets – it’s a multi-aspect thing.
The question to ask is: does the functionality match what you’re looking for? Maybe what we should be pushing for is a better understanding at the decision-making level.
“At the end of the day, customers will get what they want in terms of performance, robustness and so on but maybe not at the prices they expect to pay. Many customers have similar needs.
If we look at capturing images of licence plates or window-mounted toll stickers, for example, there are many different ways of achieving that goal; auto-brightness, multiple exposures, speed of acquisition of the camera, resolution, full-frame or part-frame scans are just some of the things that have to be considered.
The different solutions are subtly nuanced; there’s not necessarily one ‘right’ way of doing things and as all of the current solutions tend to be pretty good you won’t get a 50% increase in performance by doing things one way rather than another. But you might get tweaks which give you the edge in some circumstances.
“Perhaps if bureaucrats were less prescriptive and systems developers were given more flexibility we’d see more progress. Sometimes it’s best to step back and be more results-led, not least because it means that technology can be allowed to evolve as it needs to, not be forced down certain paths.
7094 Basler’s ITS product line manager Enzio Schneider says one of the biggest issues facing machine vision is the huge variety of camera solutions available to the market. This leaves end customers’ decision-makers (who are typically governmental officials) increasingly insecure about what the optimum fit might be for their requirements. They can choose from really quite crude analogue cameras up through slightly more sophisticated offerings, progressing to block camera solutions which include elements of machine vision and then full-blown machine vision. There are also consumer-grade cameras although these have disadvantages in terms of reliability and extremely short product cycle times.
“Machine vision’s capabilities are well regarded but have tended to be in top-end applications such as tolling and enforcement,” he states. “At the decision-maker level, a lot of awareness of feature capabilities, such as auto-iris, focus and white balance, has come from the consumer technology – as have price pressures. We end up with a situation where performance levels are misunderstood or simply not required; very high resolutions, for instance, don’t necessarily improve image quality. As a result a lot of the price:performance ratios we see in the ITS sector are not quite right.”
Any education process is going to be a product of time, he feels.
“One reason for this is the huge number and variety of players in the ITS market. I think we’re going to see a very necessary process of consolidation in the next few years as people better realise the best technologies for different applications – remember that it’s very easy to ‘prove’ performance levels on paper but it’s something quite different to deliver them in the real world.
“A major challenge for camera suppliers like Basler is that their customers aren’t the end users; they’re the solutions providers for the end user’s projects. It’s for our customers to educate the end users, as our access to them is virtually non-existent. It’s an issue which is compounded by the large regional differences in thinking in terms of operations and what’s best, so I really think that time is the answer here.”
The recent rush by some organisations within the ITS sector to associate themselves with the term ‘machine vision’ underlines just how important the technology has become in a relatively short space of time. However, despite the technology having been applied in certain traffic management applications for some years, there remains a significant shortfall of knowledge as to what machine vision is and what it can do.
While this is inevitable wherever new technology is applied, in some respects historical attitudes within the ITS sector are also responsible for the situation. Traffic management shares much with the security sector (for instance) in that it has become used to cameras being commodity items and a constant, downward pressure on price for a given level of performance. This drives a cost-consciousness which is absent from more traditional machine vision markets such as production processing, where it is more readily accepted that solutions have to be bought up to a quality and not down to a price.
Machine vision suppliers have moved to adopt the technology and address market needs - as ITS International previous detailed – although other, related issues, still remain. One is the presumption on the part of decision-makers of a camera’s capabilities based on historical norms – it is rather too easy to consider machine vision as ‘just’ an enhanced or next generation of camera, rather than allowing new ways of doing things. Another is that decision-makers can too easily decide that certain features are ‘must haves’, which then places undue restrictions on suppliers which can prevent them from engineering the most effective solutions. Yet another is caused by the complex nature of machine vision itself: specialisation at both the camera and component levels means that many of the machine vision sector’s experts actually have little opportunity to communicate or collaborate with end users.
As with so many ITS-related issues, the answer is education. The presumption is that the machine vision sector has to teach the traffic management sector about what the technology can do - but the reality is that education needs to become more of a two-way street. Traffic managers need to give some machine vision developers and suppliers a greater insight into what they do. The question arises as to just what might be the best way(s) in which to improve general levels of knowledge. As ever, a survey of the machine vision sector throws up as much agreement as discord.
“Perhaps social media can provide an infill”
If the traffic sector still retains an air of mystery as far as machine vision integrators are concerned, then the issue is even more acute for component suppliers, says Marc Damhaut, CEO of“My instinct would be to place ITS within the video surveillance segment but the reality is that for some applications ITS may in fact sit somewhere between the two,” he states.
Finding the right forums for meaningful interaction is difficult, he feels: “We’re a small company and we can’t hope to educate the whole world on, for example, our HDcctv/HD-SDI products.
Damhaut was surprised at seeing ITS customers at last year’s
“It may just be that Japan and other Far Eastern countries are more advanced in the application of the technology for traffic and transport applications. In the past we’ve seen showcases for the medical sector at VISION so it’s not beyond credibility to see something like that happening in the future for ITS.
“A better forum for machine vision/ITS would be great and is, probably, needed. Perhaps in the interim, social media, such as the various discussion groups on sites such as LinkedIn, can provide an infill.
“Don’t leave it to the last minute”
There remain some big educational gaps even within some of the larger ITS machine vision projects, according to Paul Downey, Business Development Manager with lighting specialist“Too often we see lighting systems being specified at the eleventh hour, when really they should be considered far earlier in a project. There are functional and commercial advantages to better and earlier integration; not only do you end up with a system which may cost appreciably less but you can also achieve better performance by being able to work out the interplays between cameras and lighting. We certainly need a greater awareness of capability versus expectation.”
Downey doesn’t see it as a case of laying blame at the traffic sector’s door. An element of teaching and learning is always to be expected as technology vendors release functionally superior products, he notes.
“With some of our high-tech products, as with other technology vendors, there’s an education process even within the expert community. Vendors wouldn’t be doing their jobs if they weren’t releasing products which push the boundaries – otherwise we’d just have a ‘me too’ environment.
“The point is that the differences in knowledge are always going to be more apparent when technology crosses sectors, such as from machine vision into traffic. But vendors also need to understand different vertical markets and have to ensure that benefits are appreciated in respective sectors. Forums are great in that respect, in that customers bring their needs and we bring our products and solutions. The debate is whether the forums we have at present are adequate. Whatever the forum style there should be onus on technology vendors to ensure that they explain and illustrate their products within an application/solution context.
“Realise the custom capabilities”
The transport sector has yet to fully come to terms with the ability of machine vision system manufacturers to build custom products, says Robert Tietz, director of business development with“‘Traditional’ security camera manufacturers are less able – and less willing – to customise unless large numbers of cameras are being ordered as they’re large box-shifters selling commodity products. The ability of machine vision camera manufacturers, who’re usually smaller and more nimble companies, is much greater.
“But there remain areas where the machine vision sector needs a better understanding of the transport sector’s requirements, such as changing light conditions and temperature ranges. Machine vision system manufacturers have been targeting this sector for a while now but I think they could still do with a much better understanding of what happens behind the camera – how the information it provides is used. That could allow us to come at a problem from a different direction entirely.
“Some of the desired information can seem quite prosaic, such as how cameras are mounted, but what goes on in the back office can change my perceptions quite considerably. What I want is the user perspective – what the person in the control room thinks and feels.
“We need a clean-sheet approach, not a situation where the sales and purchasing people are the only ones that speak to each other. When I try to teach about capabilities, I try to think about what a potential customer’s needs are – what are the light conditions going to be like, what’s the application? Perhaps what we need is a joint effort which includes integrators as well as manufacturers.”
“System integrators need to take a lead”
In all but a very few cases, many of the companies involved in machine vision have little contact with the end user. CMOSIS, for example, manufactures CMOS-based image sensors and has worked directly with companies such as“My experience is that people tend to buy what’s available and the situation with Gatso was exceptional. A complicating factor is that red light and speed enforcement legislation varies widely from country to country and has to be taken into account. Another factor is dynamic range, given that light conditions in traffic applications can vary so widely. It’s for the system integrators to lead any education effort, I think.”
"We need dedicated conference streams”
John Philips ofTraffic monitoring applications, he notes, tend to use commodity products and there are a lot of suppliers servicing a large number of vertical markets. The higher-end ITS applications tend to be more select as machine vision cameras can be triggered (an image capture feature which comes from manufacturing processing). They also offer very high frame rates - typically 100-200 frames per second (fps) versus CCTV’s 20-30fps - which improves the chances of a clear capture, and use uncompressed video. The last point is especially important because what might work for the human eye may be not adequate for automated algorithms.
“The electronic toll collection and licence plate markets are definitely getting the use of machine vision correct but there’s room for improvement in general traffic monitoring,” he says. “Traditional video is largely designed for human consumption but soon we’ll reach a point where we need more automated monitoring and automatic alarm generation.
The ‘Holy Grail’ is something like an Amber Alert situation where there’s been a child abducted and you’re trying to find a given vehicle which you know is out on the network right now. Throwing large numbers of people at the problem doesn’t work; using lots of people costs and humans are fallible. With machine vision the situation is rather different.
“But the machine vision industry could do better at introducing itself. There’s a perception that it’s too expensive and not economically feasible for some applications. That Amber Alert situation I described shows that cost is relative and we’re already seeing moves which will significantly shift the cost balance.
“There’s a big shift away from traditional computing architectures where processing goes on at the PC level. Embedded systems are starting to emerge which offer high processing capabilities in very small form factors which don’t need, for instance, the cooling fans and other peripherals required by PCs. That means they can be located in a roadside cabinet or up a gantry or mounted with a camera. That’s a big upward shift in intelligence at any given point on a network and big downward shift in cost - and it’s happening now. We’re already seeing these solutions in the military market and I think that inside two to three years we can expect something of an explosion in the civil sector.
“But how do we communicate all this? Chat groups, I’ve found, are a good way to get information out but they’re much better if people already know what they’re looking for. Perhaps a better way is to present at more industry-centric shows, such as those for ITS; and given the huge potential applications of machine vision, I don’t think it’s unreasonable that the technology gets its own, dedicated conference streams.”
“It’ll take time… and money”
“The requirements we see are much more fully formed. The problem over the last few years has been that there’s simply been no money to invest in new technologies or projects. However the last six months have seen some positive movement in this regard.
“We’re seeing products with broader ranges of application as a result – people are less interested in a camera that will ‘just’ do spot speed measurement, for instance.
“A huge amount of development has gone on with sensors. System suppliers have worked to overcome a range of challenges in a series of market verticals. Ultimately, the traffic sector isn’t big enough on its own to drive development but it can take advantage of what’s being done on behalf of consumer electronic and other sectors.
We’ve done the job of educating our customers – the camera companies – and I think economic upturn is going to be the driver from here. Greater awareness of machine vision’s capabilities will come from a greater number of deployments. That’ll take money and time.”
“It’s tempting to interfere”
“It’s tempting to interfere when you have a little knowledge. An example is where the desired next level of magnification is stated to be ‘X’ but there’s no realisation that increasing resolution may not always give a better result,” he says. “There’s a good grasp at the engineering level but not always at the policy or decision-making levels.
“Right now the main constraint in the ITS sector is price, not system efficiency -it should be the other way around. Price point is based on what people expect to get from a feature point of view but management may reconsider targeted price points if they’re shown the benefit of advanced feature sets – it’s a multi-aspect thing.
The question to ask is: does the functionality match what you’re looking for? Maybe what we should be pushing for is a better understanding at the decision-making level.
“At the end of the day, customers will get what they want in terms of performance, robustness and so on but maybe not at the prices they expect to pay. Many customers have similar needs.
If we look at capturing images of licence plates or window-mounted toll stickers, for example, there are many different ways of achieving that goal; auto-brightness, multiple exposures, speed of acquisition of the camera, resolution, full-frame or part-frame scans are just some of the things that have to be considered.
The different solutions are subtly nuanced; there’s not necessarily one ‘right’ way of doing things and as all of the current solutions tend to be pretty good you won’t get a 50% increase in performance by doing things one way rather than another. But you might get tweaks which give you the edge in some circumstances.
“Perhaps if bureaucrats were less prescriptive and systems developers were given more flexibility we’d see more progress. Sometimes it’s best to step back and be more results-led, not least because it means that technology can be allowed to evolve as it needs to, not be forced down certain paths.
“Time is going to be the answer”
“Machine vision’s capabilities are well regarded but have tended to be in top-end applications such as tolling and enforcement,” he states. “At the decision-maker level, a lot of awareness of feature capabilities, such as auto-iris, focus and white balance, has come from the consumer technology – as have price pressures. We end up with a situation where performance levels are misunderstood or simply not required; very high resolutions, for instance, don’t necessarily improve image quality. As a result a lot of the price:performance ratios we see in the ITS sector are not quite right.”
Any education process is going to be a product of time, he feels.
“One reason for this is the huge number and variety of players in the ITS market. I think we’re going to see a very necessary process of consolidation in the next few years as people better realise the best technologies for different applications – remember that it’s very easy to ‘prove’ performance levels on paper but it’s something quite different to deliver them in the real world.
“A major challenge for camera suppliers like Basler is that their customers aren’t the end users; they’re the solutions providers for the end user’s projects. It’s for our customers to educate the end users, as our access to them is virtually non-existent. It’s an issue which is compounded by the large regional differences in thinking in terms of operations and what’s best, so I really think that time is the answer here.”
