Search Result

Vulnerable Road Users (VRUs), such as pedestrians and cyclists, are among the most exposed participants in road traffic environments, and their reliable perception remains a key challenge for Advanced Driver Assistance Systems (ADAS) and autonomous vehicles. Automotive radar operating in the high- frequency 77 GHz band provides robust sensing capabilities with high range and angular resolution, making it well suited for VRU detection under adverse weather and lighting conditions. This paper presents a real-time artificial intelligence–based frame- work for VRU classification using 77 GHz automotive radar measurements. Radar signal processing techniques are applied to extract discriminative kinematic and micro-Doppler features that capture the distinctive motion characteristics of pedestrians and cyclists. These features are used to train a Support Vector Machine (SVM) classifier, selected for its strong generalization capability and low computational complexity, which is critical for real-time automotive applications. The proposed system is evaluated using real-world radar data collected in urban traffic scenarios, demonstrating reliable separation between pedestrian and cyclist classes across varying speeds and motion patterns. The results indicate that the proposed approach provides an effective and computationally efficient solution for real-time VRU classification, supporting its integration into safety-critical ADAS and autonomous driving systems.

Traffic congestion at urban roundabouts represents a critical challenge in developing cities such as N’Djamena, the capital of Chad, where rapid urbanization, heterogeneous traffic composition, informal driving behaviors, and limited road infrastructure significantly degrade mobility and traffic efficiency [1], [2]. In such environments, conventional fixed-time and rule-based traffic signal control systems lack the adaptability required to respond effectively to highly dynamic and uncertain traffic conditions. This paper proposes a reinforcement learning–based intelligent traffic signal control framework specifically designed for signalized roundabouts in developing-city contexts, with a particular focus on the traffic characteristics of N’Djamena. The proposed system models traffic signal control as a sequential decision-making problem, in which an autonomous reinforcement learning agent continuously observes real-time traffic states—such as queue lengths, vehicle waiting times, and traffic density—and learns optimal red-light and green-light timing policies through direct interaction with the traffic environment. Unlike traditional approaches, the proposed method does not rely on predefined signal plans or prior traffic flow models, enabling it to adapt effectively to fluctuating and unbalanced traffic demand typical of N’Djamena’s urban road network. The effectiveness of the proposed approach is evaluated through traffic simulation experiments configured to reflect realistic traffic conditions observed in N’Djamena, including heterogeneous vehicle types and variable demand patterns. Performance comparisons with conventional fixed-time signal control demonstrate substantial reductions in average vehicle waiting time, queue length, and overall congestion. These results confirm the potential of reinforcement learning–based traffic signal control as a scalable and cost-effective solution for adaptive traffic management in resource-constrained urban environments and highlight its applicability to intelligent transportation systems in developing cities.