Stereo and IMU assisted visual odometry on an OMAP3530 for small robots

Abstract

Small robots require very compact, low-power, yet high performance processors for vision-based navigation algorithms like stereo vision and visual odometry. Research on real-time implementations of these algorithms has focused on FPGAs, GPUs, ASICs, and general purpose processors, which are either too big, too hot, or too hard to program. System-on-a-chip (SoC) processors for smart phones have not been exploited yet for these functions. Here we present a real-time stereo vision system with IMU assisted visual odometry implemented on a single Texas Instruments 720Mhz/520Mhz OMAP3530 SoC. We achieve frame rates of 46 fps at QVGA or 8 fps at VGA resolutions while simultaneously tracking up to 200 features, taking full advantage of the OMAP3530's integer DSP and floating point ARM processors. This is a substantial advancement over previous work as the stereo implementation produces 146Mde/s in 2.5W, yielding a stereo energy efficiency of 58.8Mde/J, which is 3.75× better than prior DSP stereo while providing more functionality.