Robot tracking in low-power visual sensor networks

Abstract

The paper addresses the problem of visual tracking in low-power visual sensor networks. Accurate real-time localization and energy efficiency are key issues in such applications. A novel approach for tracking and identification of robots using 1D barcodes is proposed. The nodes in the network perform visual processing and export processed information in form of robot codes and motion vectors. The central server can turn on/off processing on either node based on the received information. Furthermore, a novel LoRa-based low-power communication protocol is designed to provide the necessary bandwidth. The proposed system yields optimized node energy consumption, which we model mathematically and compare against the energy consumption of a system of independent cameras. The experimental results prove the correctness of the formulated model and demonstrate the advantages of the proposed approach. The performance of the system is practically demonstrated on LoRa-enabled Raspberry Pi cameras.