%0 Journal Article
%T Object Recognition and Pose Estimation on Embedded Hardware: SURF-Based System Designs Accelerated by FPGA Logic
%A Michael Schaeferling
%A Ulrich Hornung
%A Gundolf Kiefer
%J International Journal of Reconfigurable Computing
%D 2012
%I Hindawi Publishing Corporation
%R 10.1155/2012/368351
%X State-of-the-art object recognition and pose estimation systems often utilize point feature algorithms, which in turn usually require the computing power of conventional PC hardware. In this paper, we describe two embedded systems for object detection and pose estimation using sophisticated point features. The feature detection step of the ˇ°Speeded-up Robust Features (SURF)ˇ± algorithm is accelerated by a special IP core. The first system performs object detection and is completely implemented in a single medium-size Virtex-5 FPGA. The second system is an augmented reality platform, which consists of an ARM-based microcontroller and intelligent FPGA-based cameras which support the main system. 1. Introduction Computer vision (CV) and augmented reality (AR) are growing areas of research with many applications. For example, automotive industry makes use of passive optical sensors in the field of offboard traffic observation and management [1¨C3]. Onboard systems often utilize CV techniques for driver assistance and traffic sign detection [4, 5]. Turning CV to account, AR enhances real environments by virtual elements and allows manifold applications such as guided order picking or maintenance tasks [6, 7]. Optical object detection and pose estimation are very challenging tasks since they have to deal with problems such as different views of an object, various light conditions, surface reflections, and noise caused by image sensors. Presently available algorithms such as SIFT or SURF can to some extent solve these problems as they compute so-called point features, which are invariant towards scaling and rotation [8¨C11]. However, these algorithms are computationally complex and require powerful hardware in order to operate in real time. In automotive applications and generally in the field of mobile devices, limited processing power and the demand for low battery power consumption play an important role. Hence, adopting those sophisticated point feature algorithms to mobile hardware is an ambitious, but also necessary computer engineering task. This paper describes two embedded systems for SURF-based object recognition and pose estimation. The first system performs feature-based object recognition and has been implemented as a SoC on a single FPGA (Virtex-5 FX70). It can process images at a frame rate of up to five frames per second and (in our experiments) recognize and distinguish 9 different objects at a sensitivity of 91% and a specificity of 100% (no false positives). The second system determines the 3D pose of one or more objects. It features an ARM
%U http://www.hindawi.com/journals/ijrc/2012/368351/