Edge-Enabled Collaborative Object Detection for Connected Autonomous Vehicles

• Everett Richards, Computer Science, San Diego State University
• Bipul Thapa, Computer Science, University of Delaware

• Lena Mashayekhy, Computer and Information Sciences, University of Delaware

Accurate and reliable object detection is crucial in the world of Connected Autonomous Vehicles (CAVs) to avoid collisions and ensure safety. Traditional cloud-based solutions fall short of meeting this requirement in stringent real-time autonomous driving, necessitating a shift towards decentralized edge computing. Moreover, relying on a single onboard sensor is insufficient for accurate and consistent decision-making, thus requiring a collaborative approach that integrates multiple viewpoints for real-time decision-making to ensure a robust and safe system. In this paper, we introduce an innovative framework, Edge-Enabled Collaborative Object Detection (ECOD) for CAVs, that leverages edge computing and collaborative object classification to intelligently detect objects and track parking availability in dynamic traffic scenarios, thereby enhancing CAVs’ safety and decision-making capabilities. Our ECOD framework includes two key algorithms: Perceptive Aggregation and Collaborative Estimation (PACE) and Variable Object Tally and Evaluation (VOTE). These algorithms allow CAVs to collaborate and share parking spot and object detection data with an edge server, which aggregates and processes this information to achieve consensus on object classifications. We develop a testbed consisting of camera-equipped robotic cars and an edge server and evaluate the efficacy of our framework. Our experimental results demonstrate the significant benefits of our approach in terms of improved object classification accuracy, outperforming traditional single-perspective onboard methods by as much as 75%. This research has a high potential to improve CAV safety and provide new perspectives on intelligent transportation systems.