Vehicle target detection model based on CenterNet coding optimization
In order to solve the problem of low vehicle type recognition rate in vehicle recogni-tion systems,this paper proposes a vehicle target detection model that optimizes CenterNet enco-ding.This mode1 uses ResNet18 as the basic network to ensure real-time performance.In view of the problem that simultaneous target detection and classification in the original model may lead to confusion and mutual interference of vehicle target information,this model decouples detection and classification,focusing on the inherent characteristics of vehicles and the unique characteristics between models can effectively improve the accuracy of classification.In order to solve the problem of inaccurate identification of special vehicle models,the model adds the width,height,shape and other characteristic information of the vehicle target into the classifica-tion for feature learning to improve the accuracy of special vehicle classification.Experimental results show that on the UA-DETRAC data set,the average accuracy of this model has increased by 0.5 percentage points,and the F1-Score has increased by 1.1 percentage points.
NETL
NSTL
万方数据
