An adaptive license plate recognition framework for moving vehicles

The unprecedented increase in the number of vehicles and a proportionated increase in traffic abnormalities have led to inconsistencies in traffic regulation systems, causing a detrimental impact on human safety. As a result, the speed computation of fast-approaching vehicles along with the extraction of License Plate characters has now become one of the popular areas of study. In this paper, a collaborative framework has been proposed to anticipate the speed of the fast-approaching vehicle and extract the License Plate Characters of the vehicle. The proposed system primarily comprises three successive steps, Object detection which is performed using YOLOv5 (You Only Look Once) on the processed video stream, Vehicle Speed Estimation, and License Plate Recognition (LPR). The relative speed is determined using the relative distance travelled by the vehicle over consecutive frames. An absolute error of 0.98 km/hr was found when comparing the computed and actual world speeds. We have used WPOD-NET for localizing the license plate and in the final section, we obtain the results of baseline pertained models such as-MobileNet, Xception, and ResNet50. The highest overall system accuracy for Vehicle License Plate Recognition module is 90%.