PETRI: Reducing Bandwidth Requirement in Smart Surveillance by Edge-Cloud Collaborative Adaptive Frame Clustering and Pipelined Bidirectional Tracking
Hosted in Virtual Platform
Design of Cyber-physical Systems and IoT
DescriptionIn smart surveillance, neural networks running on cloud servers require large bandwidth to upload videos. Edge-cloud collaborative encoding based on ROI (Region-Of-Interest) can reduce bandwidth requirement, but it suffers from inaccurate ROI detection due to feedback latency and undetected new targets. To address the above challenges, we propose an object detection system which adopts a latency-hiding pipeline workflow with adaptive keyframe interval selection for different input videos, and utilizes a retrotracking method to find undetected targets. With negligible average precision impact, we achieve up to 66.44% bandwidth reduction compared to cloud-only method and 30.25% bandwidth reduction compared to previous work.