Year: 2026
Type: article
Abstract: The high-speed object tracking is not only a problem of tracking fast-moving targets, but also a system-level challenge involving visual sensing, data representation, motion modeling, target association, and hardware implementation. This paper reviews object-recognition-driven methods for high-speed object tracking, with a focus on data representation, motion information extraction, event-based vision, and hardware system design. First, frame-based image sequences, optical flow, and event streams are compared in terms of their ability to represent target motion under high-speed conditions. Then, traditional motion cues, deep representation methods, and lightweight event-driven features are discussed from the perspectives of robustness, real-time performance, and deployability. The paper further analyzes the role of high-speed visual sensors, event cameras, FPGAs, and hardware–software co-design in practical tracking systems. Existing studies show that event-based vision and dedicated hardware acceleration can reduce redundant computation and improve temporal response, but stable association, resource efficiency, and real-world deployment remain open challenges. This review argues that future high-speed tracking systems should be developed through the co-design of sensors, algorithms, and hardware architectures.
DOI: 10.54097/ry9t1148
Author: Hanze Jin
Institution: RWTH Aachen University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7203479976