A Blockchain-Enabled Multiple Object Tracking for Unmanned System With Deep Hash Appearance Feature

Abstract

The unmanned system based on mobile-edge computing (MEC) and blockchain can conveniently share computing resources and realize multi-devices collaboration. However, frequent data communication in the framework of object detection brings a heavy computational burden. In this paper, a novel blockchain-based multi-view unmanned equipment fusion architecture using multiple object tracking (MOT) technique is designed. Inspired by the idea of person re-identification and hash representation for image retrieval, a novel MOT technique using HashNet to extract deep hash appearance of objects is proposed. In addition, based on the blockchain and MEC technology, we make some improvements in feature fusion and tracking interrupt recovery. We combine deep hash appearance features with motion features and design a tracking interruption recovery mechanism to solve the problem of object occlusion. The experiment results on the MOT challenge dataset demonstrate that the proposed algorithm can handle object occlusion problem effectively and successfully reduce the number of identity switches. In real application scenes our algorithm performs particularly well, showing that our algorithm is more practical.

Keywords

• Multiple object tracking
• computer vision
• blockchain
• hash representation
• person re-identification
• machine learning