Real-Time and Low-Latency Processing in Computer Vision: A Literature Review

Abstract

The capacity for computer vision (CV) systems to process visual data with
minimal delay is fundamental for applications demanding immediate interaction with dy-
namic environments. This review surveys advancements from 2020–2025 aimed at enabling
real-time, low-latency CV. We examine core techniques including hardware acceleration,
model compression strategies (quantization, pruning), parallelism, the design of efficient
network architectures, knowledge distillation, and the deployment of edge computing. Key
application domains such as autonomous driving, intelligent surveillance, augmented and
virtual reality (AR/VR), real-time video streaming, and robotics are discussed to illustrate
how these methodologies address specific latency challenges. The primary focus remains on
mitigating computational and processing latency inherent in vision tasks, deliberately ex-
cluding network-level communication optimizations. Through a structured exploration, this
review highlights recent systems and frameworks that are actively redefining the frontiers of
instantaneous visual perception and intelligent response.