Colorful Future
The search for 'human vision' in detection is a step closer with improvements in colour camera technology
Traffic engineers worldwide have been searching for a method to effectively 'see' the complete traffic scene and perform not only cost-effective loop emulation, but accurate automatic incident detection, queue length detection and traffic data collection, since the first video vehicle detection system, Autoscope from Image Sensing Systems (ISS), was developed in 1984, and introduced commercially in 1989.
Market leaders in the industry continue to listen to their customers and reinvest in technology, research and development, seeking a cost-effective means to provide clearly superior detection accuracy and image clarity with colour cameras in challenging weather and lighting conditions.
The primary task of any video vehicle detection system is to make a clear distinction between the pixel fragment values belonging to roadways and to vehicles.
In monochrome imagery, varying shades of grey represent all regions of the traffic video. It is possible that when using only monochrome video cameras, that the perceived shades of grey which represent visually in black and white, to be the same shade of grey represented by the pixel fragments of the roadway under a specific detection zone(s).
The latest developments and incorporation of both temporal and spatial machine vision techniques have dramatically enhanced the detection accuracy of video vehicle detection systems. Autoscope Solo Pro now features enhanced image processing algorithms that derived from colour cameras. The use of colour cameras for video vehicle detection can now resolve many routine anomalies found in early video vehicle detection systems, providing vehicle presence, count and speed accuracies in excess of 98%.
ISS says that since the first Autoscope system was sold, camera and processor technology has steadily advanced.
Autoscope systems already installed, and those to be installed in the future, will continue to perform accurately with monochrome cameras. Additional real-time traffic information and image detail can, for the first time, be derived from the use of colour cameras (not to mention the enhanced visual benefits from being able to view real-time colour video vehicle detection) either in the field or remotely via low-power microwave or compressed video over standard dial-up telephone lines.
Surveillance and video detection are being merged into a single application for the benefit of the traffic engineer and the safety of the motoring public.
Accompanied by major improvements in processor technology, camera technology and the degree of integration with optics have dramatically improved since the early 1990s.
It is now possible to obtain a colour camera with a lens and associated zoom and focus control in a package which is smaller than that of the original monochrome cameras used for video vehicle detection as recently as 1998. The cost of colour cameras is nearly equivalent to black and white cameras, yet they provide better visualisation even during periods with low light levels.
A monochrome picture is essentially two-dimensional, offering width and height information. If all objects in the picture are the same shade of grey, differentiation from one another is impossible, and errors in vehicle presence, speed, lane occupancy, and queue length may be experienced.
When we actually utilise the chromaticity from a traffic video image, we can now add a new dimension of depth to the image. The image is now three-dimensional-like, and it is possible to separate objects in the image that have the same grey scale level by examining their location and depth dimension. Improved separation of objects in the image and from the background reference image is important in providing enhanced detection accuracy and reliability over the broad range of challenging lighting and environmental conditions faced by modern video vehicle detection systems.
The addition of specialised algorithms to Autoscope Solo Pro improve vehicle detection and traffic data collection accuracy in the following areas:
- stopped vehicle detection;
- queue length estimation;
- rejection and elimination of static and moving shadows;
- vehicle classification based on colour, and
- rejection of headlight reflections.
Several new applications for video vehicle detection are under development, such as deriving individual travel times of vehicles. Keeping in mind our long-term goal of reducing traffic delays, travel time, congestion, fuel consumption, air pollution and enhancing motorist safety - video vehicle detection continues to evolve and play an even stronger role in typical as well as new and unique applications.
Video vehicle detection is now more cost-effective, accurate and reliable than ever, outperforming inductive loops and other above-ground detection technologies under all weather and lighting conditions. For many years, video vehicle detection or machine vision has been incorrectly referred to as "simulated human vision".
Now, with the addition of colour and the added benefits it provides to enhancing system accuracy and reliability, the technology is one step closer to meeting the ideal of "human vision". The world is made up of colour, not shades of grey, so why not benefit from the advances of technology?
Authors
Oran Ener, Richard Jacobson and Dr. Bill Sowell, Image Sensing Systems, Inc.As published in World Highways April 2001
©2001 World Highways
