Inspired by human perception, a novel framework for dynamically allocating algorithmic and computational resources to achieve variable precision tracking of extended objects is presented. Probabilistic object relevancy metrics reflect the priority of each tracked object to the consumer of the tracking output, and are leveraged to trigger mode transitions in a hybrid system implementation of the proposed priority-based framework. In this way, the bulk of the algorithmic and computational resources are reserved for tracking objects of highest priority with high-precision methods, while low priority objects are tracked with inexpensive, qualitative methods. An example implementation of the proposed framework is provided for an autonomous driving application, in which the consumer of the tracking output is an anticipatory path planner. Simulation results demonstrate the ability of the framework to automatically trade computational complexity for tracking precision as a function of an object’s priority to the tracking consumer.