The ability to recognise road and traffic signs is becoming an important research area in intelligent transport systems. It has a number of applications in driver support and highway maintenance systems; and intelligent autonomous vehicles. Fred Senekal, a senior CSIR researcher involved in intelligent field robotics studies, has recently published a research paper in which a new computer algorithm for the detection and classification of triangular traffic signs, such as warning and yield signs, is presented.
The CSIR's rover
Traffic signs are designed to have specific shapes and saturated colours that are easily distinguishable from their environment. Senekal says, "We have sought to develop a fast and robust algorithm to exploit these facts as part of a vision system for the CSIR's CAR (CSIR Autonomous Rover) project. Vision is one of the most difficult tasks to accomplish in an intelligence machine."
In South Africa and many other countries, typical control, prohibition and warning signs contain red, black and/or white. They also have specific shapes - warning signs and yield signs are triangular, for example. "Our algorithm was applied to a set of images obtained from a camera mounted on a moving vehicle. To create the template images, reference sheets of the official traffic sign designs were obtained from the Department of Transport. From these sheets, the templates for 87 warning signs and two yield signs were created. Our technique proved efficient - good detection and classification performance was achieved under normal daylight conditions in the absence of occlusions."
Intelligent transport systems could focus a driver's attention to road conditions ahead, such as pedestrians crossing the road or a change in the speed limit, allowing the driver to take appropriate action on time. In intelligent autonomous vehicles, the ability to recognise and interpret such signs could contribute greatly to their control and safe navigation. Senekal adds, "For example, a sign indicating that there is a stop ahead may lead the control system to reduce the speed of the vehicle. In highway maintenance and sign inventory applications, the ability to recognise and possibly evaluate the condition of the sign can greatly reduce the cost and effort of maintaining road infrastructure."
What lies on 'the road ahead'
Traffic authorities place traffic signs near the road surface in a clearly visible position, usually free from any occlusions. However, there are a number of difficulties that make autonomous vehicles' recognition of traffic signs difficult. The physical condition of the sign is important, given that performance may degrade when the sign is damaged, or as the quality of paint deteriorates over time. "There may be graffiti present, low-light conditions, shadows, reflections, or even partial or full occlusion by static or moving objects," explains Senekal.
"We need to extend our work to make it more robust to various environmental influences, especially the effect of partial occlusions. In addition, we would like to track a sign across multiple frames in a video sequence, which could lead to reduced computational cost and make the system operate in real time."
Senekal and his team are also working on other computer vision studies applicable to autonomous vehicles, such as the ability to analyse a video sequence to detect the road surface, lane markings and obstacles such as pedestrians and other vehicles.
Enquiries: CSIR Communication