Author(s)

Helin Zhu, Hua Pan

Corresponding Author

Helin Zhu

ABSTRACT

After the cancellation of provincial toll stations, a large amount of abnormal charging data appeared in actual applications. These data have more or less OBU information missing or even lack of image information, which cannot be accurately deducted. In view of this situation, this paper proposes a vehicle multi-dimensional feature recognition method based on cascaded convolutional neural network, which constructs the vehicle face feature information, and can subsequently restore the trajectory of abnormal vehicles based on these information. Finally, the experimental results show that the method can quickly and accurately find vehicles with abnormal OBU or suspected evasion of expenses, and identify their trajectories, thereby achieving more accurate billing.

KEYWORDS

ETC; Vehicle face recognition; Deep learning; Convolutional neural network