An Effective Lightweight Crowd Counting Method Based on an Encoder–Decoder Network for Internet of Video Things

• DOI: 10.1109/jiot.2023.3294727
• Publication Date: 2023-07-12T17:24:48Z
• Authors: Jun Yi; Fan Chen; Zhilong Shen; Yi Xiang; Shan Xiao; Wei Zhou
• Keywords: Edge device; Deep Packet Inspection

An emerging Internet of Video Things (IoVT) application, crowd counting is a computer vision task where the number heads in crowded scene estimated. In recent years, it has attracted increasing attention from academia and industry because its great potential value public safety urban planning. However, become challenge to cross gap between increasingly heavy complex network architecture widely used for pursuit with high accuracy constrained computing storage resources edge environment. To address this issue, an effective lightweight method based on encoder–decoder network, named (LEDCrowdNet), proposed achieve optimal tradeoff performance running speed applications IoVT. particular, improved MobileViT module as encoder designed extract global-local features various scales. The decoder composed adaptive multiscale large kernel (AMLKA) atrous spatial pyramid pooling process (LC-ASPP), which can perform end-to-end training obtain final density map. LEDCrowdNet suitable deployment two platforms (NVIDIA Jetson Xavier NX Coral Edge TPU) reduce floating point operations (FLOPs) without significant drop accuracy. Extensive experiments five mainstream benchmarks (ShanghaiTech Part_A/B, UCF_CC_50, UCF-QNRF, WorldExpo'10, RSOC data sets) verify correctness efficiency our method.