Activity Analytics from Fixed Cameras for Robot Path Planning
Abstract— Although mobile robots have on-board sensors to perform navigation, path planning can benefit from fixed, or infrastructure, cameras that capture activity analytics continuously and from non-robot perspectives. We describe an approach that collects statistics of motion that repeat at some period (usually a day) and perform path planning to avoid expected activity in time and space The same statistics are used to learn preferred human paths and plan robot paths on these at times of low human activity. Temporal filtering is performed in cascade to efficiently extract long- and short-term activity in two time dimensions (isochronal and chronological) for use in global and local path planning. We compare our lightweight activity detection approach to neural network object detection methods and propose an activity-gated approach that combines activity and object detection efficiently. We deployed our approach in the ROS robot software development framework by augmenting the cost map of static objects with dynamic regions determined from activity. We describe benefits and constraints of this combination.