Object tracking bot using on-board Jetson TK1: An approach to reduce communication overhead and time delay

Abstract

Image processing has become a vital part of real time applications in today's world. However, real time image processing on mobile robots, drones and/or miniature robotic cars is still a complex task due to its high computational needs and communication overheads. The computation power of currently used embedded processing boards like Raspberry Pi, BeagleBone Black is sufficient but not enough for efficient processing in real time applications. This paper represents the combination of two disciplines related to image processing: Object recognition and Object following(tracking). In this paper, image processing is implemented in order to detect a particular object and follow it using huge computation power of Jetson TK1; NVIDIA's embedded Linux development platform featuring a Tegra Kl SOC (CPU+GPU(192 cores)+ISP in a single chip). In this paper we have implemented on-board computing to reduce the communication overhead and the time delay that occurs when the processing unit is located at a remote location.