Collaborative Multi-Agent Video Fast-Forwarding

Multi-agent applications have recently gained significant popularity. In many computer vision tasks, a network of agents, such as team robots with cameras, could work collaboratively to perceive the environment for efficient and accurate situation awareness. However, these agents often limited computation, communication, storage resources. Thus, reducing resource consumption while still providing an perception becomes important goal when deploying multi-agent systems. To achieve this goal, we identify leverage overlap among different camera views in systems processing, transmission redundant/unimportant video frames. Specifically, developed two collaborative fast-forwarding frameworks distributed centralized settings, respectively. frameworks, each individual agent can selectively process or skip frames at adjustable paces based on multiple strategies via reinforcement learning. Multiple then sense either 1) consensus-based framework called DMVF that periodically updates by establishing communication consensus connected neighbors, 2) xmlns:xlink="http://www.w3.org/1999/xlink">MFFNet utilizes central controller decide collected data. We demonstrate efficacy efficiency our proposed real-world surveillance dataset VideoWeb new simulated driving CarlaSim, through extensive simulations deployment embedded platform TCP communication. show compared other approaches literature, better coverage frames, significantly number processed agent.