Incremental Learning of Deep Neural Network for Robust Vehicle Classification

Existing single-lane free flow (SLFF) tolling systems either heavily rely on contact-based treadle sensor to detect the number of vehicle wheels or manual operator classify vehicles. While former is susceptible high maintenance cost due wear and tear, latter prone human error. This paper proposes a vision-based solution SLFF classification by adapting state-of-the-art object detection model as backbone proposed framework an incremental training scheme train our VehicleDetNet in continual manner cater challenging problem continuous growing dataset real-world environment. It involved four experiment set-ups where first stage CUTe datasets. utilized for detection, it presents anchorless network which enable elimination bounding boxes candidates’ anchors. The vehicles performed detecting vehicle’s location inferring class. We augment with wheel detector enumerator add more robustness, showing improved performance. method was evaluated live collected from Gombak toll plaza at Kuala Lumpur-Karak Expressway. results show that within two months observation, mean accuracy increases 87.3 % 99.07 %, shows efficacy method.