Architectures for FPGA-Based Implementation of Motion Estimation of Dynamic Obstacles for Autonomous Robot Navigation

Real time implementation is the key challenge in vision-based detection of dynamic obstacles in autonomous robot navigation systems. The Discrete Cosine Transform-Based Motion Estimation (DXT-ME) Algorithm, proposed for video coding has inherently parallel, low complexity computations and therefore, can be implemented in real time. It provides the exact displacement of the object of interest, making it suitable for fine-grained tracking. As two dimensional cosine/sine transforms are the main computations involved, this paper presents a comparison of three architectures for the FPGA implementation of the same. An architecture for DXT-ME is proposed.