cs.CVOct 28, 2026

Vision-Based Density Estimation for Adaptive Traffic Control

Authors: Mohammad Usaid, Deepak Silaych, Hrithik Mhatre, Archak Mittal

Organizations: Department of Civil Engineering, Indian Institute of Technology Bombay · Department of Computer Science and Engineering, University of California San Diego

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Traffic congestion is becoming a critical challenge in rapidly growing urban centers, particularly in developing nations where rapid urbanization and inadequate transport infrastructure place increasing strain on transport networks. In highly heterogeneous and chaotic traffic flows, camera-based traffic-state estimation is limited by severe vehicle occlusion and overlap. In this paper, we propose a framework to optimize traffic flow directly from stochastic visual observations. The architecture is structurally decoupled into three primary components: (a) vehicle density estimation, (b) predictive dynamics modeling, and (c) policy optimization. The proposed methodology introduces a convolutional neural network (CNN) that learns the mapping from camera feeds to spatial density maps, utilizing an adaptive k-nearest-neighbor (k-NN) Gaussian kernel to handle varying congestion levels. This feeds into an ST3DNet predictive module for spatiotemporal forecasting, followed by a hierarchical Graph-Transformer Multi-Agent Reinforcement Learning (MARL) algorithm to optimize signal phases. The density estimation stage was evaluated on 40 annotated intersection images from Rambagh 4, Jaipur, India, with a reported five-fold validation mean absolute error (MAE) of 3.32±0.953.32 \pm 0.95 vehicles and a mean absolute percentage error (MAPE) of 14.5%. These results pertain to the perception component; empirical training and evaluation of the forecasting and control stages remain future work.

Figures & tables

Figure 1: Three-stage pipeline. Current density and forecast density enter the controller state; the controller selects traffic-signal phases.
Figure 2: Example camera frame (A), annotation-derived density map (B), and YOLO26 detections (C). This illustration is not a benchmark comparison.
Figure 3: Vehicle density estimator: convolutional stem, two residual blocks, and a 32→64→1000→400→132\rightarrow 64\rightarrow 1000\rightarrow 400\rightarrow 1 regression head.
Figure 4: ST3DNet-based forecasting architecture with closeness and weekly branches, spatial recalibration, and parametric fusion.
Figure 5: Hierarchical signal controller with sub-node encoding, set-based aggregation, GATv2 and GRU processing, and phase-policy and state-value heads.
ConfigurationMean Fold MAEEnsemble MAEEnsemble MAPEMax Error
Round 1 (single model, fixed σ\sigma )2.84———
Round 2 (5-fold, fixed σ=1.5\sigma=1.5 )3.07 ±\pm 1.031.839.2%8.7
Round 3 (5-fold, adaptive σ\sigma )3.32 ±\pm 0.951.446.7%5.1
Table 1: 5-fold cross-validation results across three training rounds on 40 real-world intersection images (vehicle counts 12–49). Mean Fold MAE is the mean validation MAE across five folds. The Ensemble columns are computed over the full 40-sample dataset, where four of the five ensemble members have seen each image during training, and are therefore in-sample rather than held-out.
FoldBest MAEBest MAPEEarly Stop Epoch
Fold 03.6210.8%252
Fold 12.007.7%158
Fold 22.6713.8%404
Fold 33.4818.5%251
Fold 44.8221.5%203
Mean3.32 ±\pm 0.9514.5%—
Table 2: Per-fold breakdown for Round 3.
More in this topic

Traffic Flow Estimation

Traffic flow estimation infers and predicts vehicle volumes, speeds, densities, and origin–destination demand from incomplete transportation data.

See all 52 papers

Explore similar work

CardsList

  1. Real-Time Traffic Density Estimation: A Hybrid Computer Vision and Predictive Machine Learning Pipeline

    Oct 9, 2026Prince Kwabena Appiah BoaduDensity EstimationTraffic Flow Estimation

    Department of Computer Engineering, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana

  2. BEVLight: Generalizable Traffic Signal Control from UAV Bird's-Eye Views

    Oct 29, 2026Maonan Wang, Yuxin Cai, Junming Liu +5Traffic Signal Control

    The Chinese University of Hong Kong, Shenzhen · Shanghai AI Lab · Nanyang Technological University +1

  3. VLALight: A Vision-Language-Action Model for Traffic Signal Control

    Sep 29, 2026Pan Zhang, Siqi Lai, Kemu Dong +1Intelligent Transportation SystemsVision-Language-Action Models

    The Hong Kong University of Science and Technology (Guangzhou) · Dalian University of Technology