U27: Real-Time Commercial Vehicle Safety and Security Monitoring

摘要

Accurate real-time vehicle tracking has a wide range of applications including fleet management, drug/speed/law enforcement, transportation planning, traffic safety, air quality, electronic tolling, and national security. While many alternative tracking technologies have been developed in the recent years, license plate recognition (LPR) technology is still the simplest and readily available means for uniquely identifying vehicles in more circumstances. LPR technology has gone through quite a bit of research and development since the 1970s. While the video-imaging based approach was novel, promising, and even effective for license plate identification in the early days of the technology, accuracy of the technology has not improved much largely because of the thousands of different designs of license plates in the U.S. This study takes an entirely different path with LPR. Realizing even when a plate is read incorrectly, certain amount of useful information may still exist in the misread result. For example, when a plate of ABC 123 is read as, say, ABC I23, instead of just admitting defeat, we recognize that: 5/6 of characters were read correctly; the incorrectly read character is very similar to the correct; character, the sequence of the characters are in the right order; the number of characters are correct, etc. By using this information cleverly, one can address the plate-matching and vehicle-tracking problem with much better results. In real-time operational condition, one would not know if a plate reading of ABC I23 is correct or not and if not which character or characters were incorrect. So a challenging and complex mathematical challenge ensues here. This study boils down to the development of an algorithm for solving this problem and, for this phase particularly, demonstrating the algorithm in the field.