Perception is a critical computational task in autonomous vehicles.  Autonomous vehicles place stringent and somewhat conflicting demands on perception systems: high accuracy, low latency, and performance on limited computational resources.  The conflict between these requirements is particularly acute in the case of unmanned aerial vehicles (UAVs).  This talk will describe two related efforts on efficiency and accuracy in perception for autonomous vehicles. Work with Deep Samal, Thomas Walton, and Dung Tran of UNL builds a Markovian model for perception errors and shows its effect on tracking. Work with Krishna Muuva, Kunjan Joseph, and Santosh Pitla of UNL looks at UAV-UAV tracking.  We show that effects of latency and accuracy in the identification of target location vary depending on the current flight characteristics: in some cases latency is more important, in others accuracy. We develop a hybrid control system with control laws for both perception management and vehicle control. The perception management system selects among two different perception systems based on the relative importance of accuracy and latency.