Vehicle Position Estimation and Vehicle Classification Using Deep Convolutional Neural Networks

The aim of this paper is to classify the vehicles and estimate position with license plate localization using deep convolutional Neural Network (DCNN). Vehicle pose estimation serves as one most widely-used real-world applications in fields like toll control, traffic scene analysis, suspected vehicle tracking. We proposed a one-stage anchor-free classifier for simultaneously localizing region plates vehicles’ poses. classifier, rather than bounding rectangles, gives quadrilaterals, which more precise indication localization. For single scale input, we reached mean Precision Accuracy mAP/mAP50 35.4/82.3 on LISA benchmark dataset, already outperformed existing commercial systems OpenALPR Sighthound. multi-scale best 40.8/90.1. (front-rear), classification accuracy 98.8%, average IoU 71.3%, giving promising result an end-to-end contextual information. work has performed python programming language several libraries learning were being used purpose. Our DCNN model training started from initial weight had trained about 110000 iterations without head, so total will be around 780000 including transfer part DCNN. Transfer made start at smart point it easier optimize all functional heads simultaneously.