How do you think the new GigE standards will influence the machine vision industry?
Respond or ask your question now!
By Lee J. Nelson
Live video confers enormous advantage in the field of expressway incident detection. When an episode occurs, operators can see what is happening (type), where (location), how many vehicles are involved (severity), whether there are casualties and the overall situation. Merging collateral data with imagery facilitates alarm verification. And, swifter discovery leads to faster intervention, avoidance of secondary collisions and consequential lane closures and delays.
Technical approaches to automatic incident detection are divided into two basic categories: those employing a single sensor (station) versus comparing readings from two or more spatially separated stations. While the comparison method usually is preferred over a single sensor because the latter tends toward excessive false alarms, the former depends on inter-station communication, increases cost and may reduce reliability. Algorithms which buttress incident detection are grouped into four general classes.
Comparative Algorithms attempt to recognize and differentiate unusual traffic patterns – stopped, slow-moving or wrong-way vehicles or dropped cargo – from normal conditions. They are founded on the principle that any incident causes an increase in detector reporting levels upstream and a simultaneous downstream decrease. By evaluating certain parameters (volume, occupancy or speed) against pre-selected norms, an alarm is triggered when a measured value exceeds the preset threshold.
As the name implies, Statistical Algorithms utilize mathematical techniques to determine whether observed data diverge from estimated or predicted values. For example, lane occupancy mean and standard deviation can be calculated at one-minute intervals. An incident would be declared if current figures differ considerably from those for two prior, successive time periods or when they surpass a given setting.
Time Series/Filtering Algorithms analyze – or smooth – raw data over time to eliminate short-duration disturbances such as random fluctuations, traffic pulses and compression waves. Processed data are compared to forecasted values. When the discrepancy is significant, an incident is signaled.