Autonomous Traffic Sign Recognition and Adaptive Signal Control for Smart Cities Using Deep Reinforcement Learning

Project Abstract

This study presents an integrated framework that combines autonomous traffic sign recognition with adaptive signal control in urban environments through deep reinforcement learning (DRL). The core objective is to develop a data-driven system that enhances traffic safety and efficiency by (i) accurately detecting and classifying traffic signs in diverse real-world conditions, including occlusions, lighting variations, and weather disturbances, and (ii) dynamically adjusting traffic signal timings to optimize flow, reduce average travel time, and mitigate congestion while prioritizing pedestrian safety. The traffic sign recognition component leverages a convolutional neural network (CNN) backbone with attention mechanisms and multi-scale feature fusion, trained on a comprehensive, diverse dataset augmented with synthetic variations to improve robustness. Transfer learning and domain adaptation techniques are employed to bridge the gap between laboratory and field data, ensuring reliable performance in unseen environments. The signal control module utilizes a DRL agent based on an actor-critic architecture with a proximal policy optimization (PPO) algorithm, integrating real-time traffic state information, sign-derived constraints, and historical signal level data. The state representation combines queue lengths, vehicle speeds, pedestrian counts, signal phase, and the confidence scores from the perception subsystem, enabling the agent to learn policies that balance throughput, delay, and safety metrics under varying demand scenarios. A hierarchical control scheme coordinates local intersection decisions with network-wide optimization to achieve scalable deployment in smart city infrastructures. The research introduces a novel reward shaping strategy that penalizes unsafe or inefficient decisions while rewarding reductions in travel time, stops, and emissions, thereby aligning with sustainability goals. To evaluate performance, a multi-criteria experimental framework is employed, including simulations using a high-fidelity traffic microsimulator and a field testbed comprising equipped intersections with camera sensors and adaptive controllers. Benchmarking against conventional fixed-time and actuated control methods as well as recent DRL-based approaches demonstrates significant improvements in average intersection delay, queue length, and throughput, with notable gains in safety indicators such as near-miss events and sign recognition accuracy under adverse conditions. Sensitivity analyses examine the impact of perception errors, communication latency, and environmental variability on overall system performance. The study also explores transferability across intersections of different geometries and traffic patterns through a modular design and standardized interfaces, enabling plug-and-play deployment in existing smart city ecosystems. Findings indicate that integrating robust traffic sign understanding with adaptive, learning-based signal control can yield synergistic effects, where accurate sign detection informs appropriate pedestrian and vehicle prioritization, and optimized signal timing, in turn, enhances recognition reliability by providing clearer contexts for perception modules. The research contributes a comprehensive architecture, algorithmic breakthroughs, and empirical evidence supporting the practicality and benefits of deploying DRL-enabled perception and control in urban traffic management.

Project Overview

What This Project Is About

---

A beginner-friendly overview of developing a system that recognizes traffic signs and adjusts traffic lights in real time to improve city traffic flow, using basic learning techniques.

---

---

The Problem It Addresses

---

Cities struggle with unpredictable traffic patterns and outdated signaling. Recognizing signs reliably helps ensure correct, timely signal changes. The project aims to combine sign recognition with smart signal control to reduce delays and improve safety.

---

---

Objectives of the Project

---

1. Understand how traffic signs are detected from camera data.
2. Learn how signals can adapt based on traffic conditions.
3. Build a simple, safe prototype that demonstrates the idea.
4. Evaluate improvements in travel time and waiting periods.

---

---

What You Will Do Step by Step

---

1) Study basic ideas behind image recognition and traffic signal logic. 2) Collect or use existing traffic sign images and synthetic traffic data. 3) Create a simple model to identify signs from images. 4) Design a basic rule or learning method to adjust signals. 5) Test the system in a simulated city and measure results. 6) Analyze what worked and what didn’t, and suggest improvements.

---

---

Expected Outcome

A working concept that shows how traffic signs can be recognized and signals adapted to reduce delays, with a clear explanation of potential real-world impact and next steps for further research.