Real-Time Traffic Congestion Prediction Using Multi-Stage Image Enhancement and Dynamic Weighted Object Count Estimation

Abstract

The research is a structural analysis of traffic capable of offering support to the validity of real-time surveillance during severe weather conditions. The modern traffic monitoring systems are usually prone to noise interference, segmentation error and poor consistency when tracking vehicles and this leads to poor estimation of congestion. The general purpose of the work is to have a robust pipeline that operates through several stages to generate high-clarity denoising, proper segmentation, robust classification and dynamic density estimation. The methodology is supplemented with adaptive frequency-selective noise suppress (AFSNS) to enhance the quality of the image and then Context-Aware Region-Aggregated Segmentation (CARAS) is applied to isolate the vehicle regions that are structurally consistent. Vehicle Enhanced Discriminator and Classifier Network (VEDCN) improves the multi-class recognition and Dynamic Weighted Object Count Estimation (DWOCE) computes the lane density using weighted vehicle density and motions. The experimental findings indicate that the denoising, segmentation as well as recognition performance have increased to a great extent where denoising attains 92.39 percent denoising accuracy, segmentation attains 92.97 percent segmentation accuracy and general density estimation attains 95.6579 percent accuracy. The hybrid system provides enhanced clarity, structural information is maintained and also becomes consistent with changing lighting and traffic conditions. In conclusion, the proposed framework offers flexible, high accuracy, and scalable solution to the intelligent transportation systems in such a way that the real-time tracking and forecasting of the congestion would be more precise in the intricate road environment.