An Energy-Efficient Kalman Filter Coprocessor Design for Multiple-Object Tracking
An Energy-Efficient Kalman Filter Coprocessor Design for Multiple-Object Tracking Targeting at Video Understanding
Abstract:
Energy-efficient Kalman filter (KF) designs are inevitable in multiple-object tracking (MOT), a challenging computer vision task, for real-time mobile applications such as smart glasses, nanodrones, and mobile robots. VLSI design for KF is a feasible way to implement the required complex matrix operations; however, direct implementations are inefficient due to the data sparsity and redundant calculations existing inside. This article introduces an energy-efficient KF coprocessor, which improves computational efficiency by employing KF formula simplifications, matrix sparsity and symmetric properties, and common coefficient sharing, along with a dedicated but flexible hardware architecture. In contrast to state-of-the-art (SOTA) application-specific integrated circuit (ASIC) designs that rigidly customize several hardware modules for specific formulations in KF, our architecture executes all formulations in one hardware, enabling both flexibility and high efficiency. Implementation results demonstrate up to 44.8× improvement in energy efficiency over SOTA solutions, achieving 110.2 GOPs/J when executed at 180 MHz, which supports 120-FPS real-time multiobject tracking.
” Thanks for Visit this project Pages – Register This Project and Buy soon with Novelty “
An Energy-Efficient Kalman Filter Coprocessor Design for Multiple-Object Tracking Targeting at Video Understanding
