Real-Time Vehicle Identification Performance Using FPGA Correlator Hardware

A real-time automatic vehicle correlator system that is capable of identifying the make, model, and series of road traffic, e.g., a Volkswagen Golf Mark 4, has been implemented in field-programmable gate array (FPGA) hardware housed on a peripheral component interconnect (PCI)-format electronic board. The system is compatible with standard desktop personal computers and accepts imagery from networked cameras. The vehicle correlation operates in conjunction with an automatic number plate recognition (ANPR) system, which acts as a coarse filter and determines the presence of a new vehicle to be identified. When the vehicle correlator is presented with a new image by the ANPR system, it correlates this image against a large database of many thousands of reference images, which represent the many vehicle variants typically found in that regional jurisdiction. The FPGA vehicle correlator can perform 10 000 correlations per second, allowing timely vehicle identification in less than 1 s when comparing against 8192 reference vehicle images. In a large-scale trial of 991 input vehicles, the vehicle correlator correctly identified 84% of the test images using a database of 62 different vehicle variants.