Considerations when using machine vision cameras in traffic applications – ITS

The latest machine vision camera technology can allow for improvements in intelligent traffic system (ITS) performance to minimize detection time in case of accidents, increase the detection rates in cases of violations, and minimize overall costs from police time and traffic jams.  With decreasing budgets, these systems also need to maintain reasonable costs and prove their efficacy to gain adoption by government decision makers. 

While some differences between traffic systems and machine vision systems are obvious, it is worth highlighting the differences for ITS designers to be aware of when considering using machine vision cameras.  Because of these differences, general purpose machine vision cameras may not be adequate, but high performance machine vision cameras are optimal for demanding applications, such as Automatic License Plate Recognition ALPR.

Traffic systems have uncontrolled lighting conditions while predictable in machine vision

With machine vision systems, there is typically controlled lighting.  Even if the lighting is limited, it is specified and predictable. 

For outdoor traffic systems, the changing lighting conditions are a major obstacle for obtaining a high quality image.  At certain times of the day the sun can be directly aligned with the camera causing blooming and smear (streaks) artifacts in the image.  Reflections from the sun on the road and headlights can also create these artifacts.  Image processing in the system cannot correct these defects so blooming and smear must be managed in the camera to ensure that the license plate is not obscured in the original image data.

Figure 1.  Example of Blooming and Smear

Often an IR flash light is used to illuminate the license plate since it is not visible for people and the license plate material is reflective at 800nm.  Starting with cameras that have significant sensitivity at this wavelength allows for the desired image quality without a complicated and costly work around.  

Constant versus continuously changing environmental conditions

While cameras within a machine vision system may experience some temperature and humidity variations, the range is usually limited.  Cameras for traffic applications are often in a protective housing, yet they must still be robust enough to handle huge temperature fluctuations and humidity.  Hot temperatures are an expected problem, but colder temperatures can also be a detrimental and cause service issues.  Lightning striking any portion of the system is another concern. With any system, component breakdown and replacement is costly, but for traffic systems particularly those over the road, the expense can be extreme if the road needs to be closed. As a minimum, industrial grade components rather than commercial grade should be used in the camera to guarantee full performance up to 50°C and down to -20°C rather than only down to 0°C.  High-performance cameras have a more robust design and specialized functionality can further enhance management of extreme outdoor conditions.

Non-precise timing and total system considerations

For inspection and metrology systems, the appearance of the device or part to test is exactly timed allowing for precise image acquisition algorithms and data management.  With license plate recognition of cars on the roads, there is an unexpected appearance of the “car” to measure, which can result in a lot of unnecessary data.  Implementation of certain processing functions on the camera such as region of interest can manage the data and significantly minimize costs elsewhere in the imaging system.

Other system differences to consider include interfaces that can safely and reliably handle the long distances between camera and PC for typical traffic system setups.    And more…

Monochrome or color cameras?

One aspect that machine vision cameras and traffic might have in common is that for high-end systems it is not the human eye that will detect the fault in the production line or the violation in traffic. It is done with high quality automated detection algorithms that most often use monochrome cameras because of their higher sensitivity and performance.

Conclusion

License plate recognition requires a crisp, clear image of a license plate to withstand dispute upon issuance of any penalties.  This means obtaining an image whenever a car appears, regardless of lighting conditions, weather conditions, the color of the license plate, or the car speed.

Unlike machine vision applications, traffic systems need to manage variable lighting conditions, unpredictable appearance and moving of “parts”, long distances to the processing system, and changing weather conditions.  Consequently, cheap general purpose machine vision cameras are not ideal for many of these systems, but high-performance cameras are an excellent option to increase accuracy and reduce overall system costs.