Transportation applications move to machine vision’s mainstream

The adaptation of machine vision to transport applications continues apace.

That the machine vision industry is taking traffic installations seriously is evident by the amount of hardware and software products tailor-made for ITS applications that are now available on the market. A good example comes from US-based 8097 Gridsmart Technologies which has developed a single wire fisheye camera that provides a horizon to horizon view for use at intersections. Not only does the single camera replace four or more in a ‘traditional’ configuration, it also covers the centre of the intersection where the traffic flows intersect.

Beyond the standard traffic management and incident detection, the horizon to horizon view gives users additional data collection capabilities as it tracks objects through the intersection and the system reports in which direction vehicles and cyclist head.  The company’s vision processing algorithm automatically counts vehicles and measures intersection performance 24/7. According to the company, such long-term counts provide a more accurate understanding of the intersection and how it is affected by seasonal variation, weather and other factors that will not be apparent in the traditional 24-48 hour studies.

Another example is the 2215 Axis P3905-RE network camera range which is designed for onboard video surveillance in or on buses, trams, subway, emergency vehicles, cars and other vehicles. In addition to the network connectivity, the camera has edge storage capabilities and an active tampering alarm function for the detection of tampering attempts such as blocking or spray-painting.

The standard 6mm lens provides a 55° horizontal field of view, while fitting the 3.6mm version widens this to 87°.

Another US company, 8142 Eutecus, that provides video analytics solutions for applications including intelligent lighting and advanced driver assistance technology, was approached by a user wanting more cost-effective red light violation detection with significantly improved performance. The unnamed user operates red light violation and speeding detection systems, traffic surveillance and other video- and radar-based solutions for municipalities across the U.S.

Video content for typical red light violation detection systems is transmitted from the intersection to a centralised facility for time-consuming manual review by operators to distinguish violators from the innocent. 

Eutecus used a Camera Link version of 593 Teledyne DALSA’s Genie TS cameras with its own Bi-i Smart Cube platform to enable it to run its Multi-core Video Analytics Engine (MVE) on the video stream.  The Genie cameras offer high-quality resolution, high frame rate, large pixels and in this case the required Camera Link interface. While the camera continuously acquires frames, the data is analysed in real-time in order that only the violations are captured and images in which no violation occurs are discarded.

In this application, the camera provides the primary video analytics source, monitoring both the traffic light and the traffic itself.

Andrew Herson, senior manager at Eutecus says: “The Genie camera essentially ‘maps’ distinct images of the traffic light as well as images of the vehicle in relation to the intersection stop line. This scene analysis is combined with data from radar, and a second, low-resolution video camera provides documentation that can be used for verification in the event a driver disputes a ticket.”

According to Euctecus, the system only sends pictures to the operator if the images of the traffic light and vehicle/stop line don’t align as they should - thereby indicating a violation. The operator then confirms the violation and generates a ticket without the need for a lengthy or exhaustive video review.

It says the automated nature of the solution eliminates the potential for human error and increases overall system reliability, and adds that it can capture and process multiple simultaneous events, such as two cars running a red light at the same time.

Streamlined solutions

As the system’s ‘intelligence’ is embedded with the camera, the entire solution can be installed above ground which Herson says provides municipalities with a streamlined and far less costly solution than traditional traffic monitoring applications.  Beyond red light violation, the system is said to be suitable for applications including toll enforcement, speeding detection, wrong-way driver detection, stopped-vehicle detection and as a traffic light sensor. 604 Vysionics is using a combination of machine vision and radar detection in its latest Victor LX system which has just been type-approved for level crossing enforcement. A scanning radar detects and tracks vehicles, cyclists and people, travelling in either direction and as they cross the stop line the camera is triggered. The machine vision system then detects if the warning lights on the level crossing signs are flashing. If so, an offence has been committed and the images are recorded, along with ANPR where appropriate, and if not they are deleted.

Geoff Collings, Vysionic’s sales and marketing director said: “Using the machine vision to detect the flashing red light on the signs makes our system non-invasive, so we do not have to break into the railway’s safety critical signalling system.”
Beyond capturing offences, the radar part of the system can gather blind data including counting vehicles, cyclists and pedestrians and classifying vehicle type into car, van and truck.

Data transmission

While machine vision helps reduce the amount of data being transferred over the network, the ever-growing number of installations means capacity limitation remains a concern.

However, the latest CoaXPress standard for HD-SDI, USB, GigE or Camera Link, defines a high-speed interface between a camera and a processing device, such as a frame grabber, and offers better performance than traditional analogue.

In its simplest form, a single coaxial cable is sufficient to transmit data at up to 625 megabytes per second, to simultaneously transmit control data and triggers, and to provide at least 13W of power to the camera. Where even higher data transmission rates are required, four parallel cables can be used to provide rates of up to 2,500 megabytes per second.

According to Jean-Bernard De Bal, vice president of business development at image acquisition and software supplier 7084 Euresys, this meets the needs of bandwidth-hungry applications. These include high speed traffic cameras shooting 560 frames per second (fps) in real-time with 4 Mega pixel (MP) resolution or ultrahigh definition 26MP cameras shooting at 80fps where the sharper images will enhance license plate recognition accuracy. Such specifications are far better than traditional full HD 1080p cameras, which shoot 2MP at 60fps.

Signal latency is another critical consideration for vehicle tracking and highway control applications and here again the CoaXPress standard can be useful as it allows the transfer of uncompressed video so zero latency is possible. High frequency real-time triggering and exposure time adjustment can also be accommodated.

CoaXPress connections can be made using the same heavy-duty coaxial cables used for analogue and HD-SDI cameras allowing analogue cameras to be replaced with high resolution digital units using the same cable network, making upgrades cheaper and easier. And as the thick copper cable inside coaxial cables is protected from electromagnetic interferences by a Faraday cage, they perform in the harshest environments and are able to withstand shocks and vibrations.

The CoaXPress standard also allows higher cable lengths compared to standards such as Camera Link or USB and lengths of greater than 100m may be achieved at half speed and greater than 40m at full speed.

While the latest software, cameras and networking may better suit traffic applications, the basis of everything remains the image sensor. Earlier in the year the machine vision sector was set unsettled by 576 Sony’s plans, reported by customers such as Framos and Allied Vision, to stop producing Charge Coupled Device (CCD) sensors in March 2017. Later Sony issued a clarification saying that CCD sensors will be available until 2026, but it does re-open the debate about the relative merits of CCD and CMOS (Complementary Metal Oxide Semiconductor) sensors.

While many favour CCDs for noise-free and geometrically accurate image of moving  vehicles needed by the ANPR algorithms, according to sensor manufacturer 8154 e2v, the introduction of global shutters has overcome the problems moving traffic posed by the original CMOS sensor’s rolling shutter.

Another reality of traffic application is the need to perform tasks during daylight, low sun, darkness and in extreme climatic conditions. Beyond reliable ANPR, the authorities also demands colour information and sometimes recognisable images of the driver and for those reasons many manufacturers  including e2v offer both colour and monochrome versions of their sensors – the latter with infrared wavelength capabilities.

Although over recent years CMOS sensors have been taking an increasing share of the traffic camera market, Sony’s plans to cease CCD sensor production (others may follow) may cause concern in some authorities. The advice from 518 Allied Vision Technologies Michael Cyros is to evaluate the expected lifetime of systems in place using cameras with Sony’s CCD sensor, and with new installations it is always advisable to check future availability of the sensors - both CCD and CMOS.

Imaging Development Systems (7178 IDS) is offering an alternative to flash or high-speed cameras with the multi integration capability of its CP, ML and LE families of cameras. In multi-integration mode, the CMOS sensor is exposed several times with the multiple integration periods as short as 100ns, allowing images of fast moving objects to be captured even in critical lighting conditions.

The number and duration of the integration periods can be set by hardware or software and according to the company the resulting blurred image can be reconstructed to allow detail such as number plate information to be obtained. In addition, image processing can be used to calculate the speed of the object and whether it speeds up or slows down by taking into account the number and length of the integration periods and the distance covered across the sensor.