Pixels to Signals: A Real-Time Framework for Traffic Demand Estimation

TL;DR

This work tackles real-time traffic demand estimation in rapidly growing urban areas by presenting a vehicle-detection framework that relies on background reconstruction from sampled frames followed by DBSCAN-based clustering of the foreground. The method computes a static background via pixel-wise frame averaging, isolates moving objects through background differencing and image preprocessing, and then clusters foreground pixels to count vehicles, achieving strong performance in distant regions. Compared against YOLOv5 and K-means baselines, the approach offers robust background quality, reduced false positives, and superior runtime efficiency, making it well-suited for deployment in environments with limited infrastructure. The study highlights practical impact for traffic signal optimization, as accurate, low-latency vehicle counts enable responsive demand estimation and improved urban mobility.

Abstract

Traffic congestion is becoming a challenge in the rapidly growing urban cities, resulting in increasing delays and inefficiencies within urban transportation systems. To address this issue a comprehensive methodology is designed to optimize traffic flow and minimize delays. The framework is structured with three primary components: (a) vehicle detection, (b) traffic prediction, and (c) traffic signal optimization. This paper presents the first component, vehicle detection. The methodology involves analyzing multiple sequential frames from a camera feed to compute the background, i.e. the underlying roadway, by averaging pixel values over time. The computed background is then utilized to extract the foreground, where the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm is applied to detect vehicles. With its computational efficiency and minimal infrastructure modification requirements, the proposed methodology offers a practical and scalable solution for real-world deployment.

Figures (8)

• Figure 1: Background reconstruction: (a) Sampled frames $F = \{f_1, f_2, \ldots, f_n\}$ at regular intervals $\Delta t$, and (b) the computed background $f_{bg}$ generated through pixel-wise mean of all frames. The operation $f_{bg} = \frac{1}{n}\sum_{i=1}^n f_i$ produces a static representation where transient objects are suppressed.
• Figure 2: Foreground isolation: (a) Original frame $f_i$; (b) Background model $f_{bg}$; (c) Isolated foreground $\mathcal{D}(f_i) = |f_i - f_{bg}|$
• Figure 3: Image preprocessing stages: (a) Grayscale and binarization ($\mathcal{G} \rightarrow \mathcal{B}$) to highlight potential vehicle regions (b) Morphological refinement ($\mathcal{M} = \mathcal{\oplus} \circ \mathcal{\ominus}$) to refine the regions and reduce noise
• Figure 4: Clusters are generated using the DBSCAN clustering algorithm, highlighting distinct groupings of data points corresponding to detected vehicles.
• Figure 5: Comparison of extracted backgrounds: (a) Proposed method and (b) K-means. The zoomed-in view at the bottom clearly illustrates that the proposed method provides a clearer background.