Vehicle Density Detection Using Hybrid SSD-Yolo-V4 Model

Abstract

Vehicle traffic congestion is a serious problem in the present day. An intriguing field of study, traffic density estimate is utilized to regulate traffic light systems for more efficient traffic management. The low data resolution and extensive vehicle identification both make this endeavor a more challenging task. This article proposes a vehicle density detection technique using a hybrid deep learning model. The selected data frames from vehicle dataset are denoised using non-adaptive threshold approach. Denoised pictures are segmented using an upgraded Prewitt algorithm and then enhanced using the CLAHE (Contrast Limited Adaptive Histogram Equalization) approach. Then the important features are selected with the use of the Convolutional neural network (CNN) algorithm. The dataset is trained with Deep Q-Learning (DQN) with Improved Spatial Pyramid neural network (I-SPPNET) architecture. Finally, vehicle density identification is done using Improved Single Shot Detector (SSD) with You Only Look Once (YOLOv4) version 4 model and the obtained results proved to be significantly better than CNN, CNN with I-SPPNET and YOVOv3.