Compressed Domain Consistent Motion Based Frame Scoring for IoT Edge Surveillance Videos

EasyChair Preprint 6764

Abstract

IoT Edge is a major active technical front. Among others, video surveillance is one of the most common use cases for IoT Edge. However, there is a need to analyze the surveillance stream, due to the sheer size of the generated data. The stream can be analyzed in either the Edge environment itself or send to a cloud for analysis. There are two main constraints to be considered. First is the associated bandwidth cost to send the data to a cloud server for processing. It underlines the need to reduce the data sent to a cloud. Second is the computational and memory constraints of the devices in the IoT Edge Environment. It implies that only computationally cheap and fast algorithms can be allowed to run in the Edge Environment. However, generally highly effective algorithms require more computational resources and memory. Pruning of uninterested frames is a viable methodology that can potentially reduce the bandwidth cost and the resources utilized. We have developed a fast, computationally cheap, and effective frame scoring algorithm that scores frames based on the consistent motion. The algorithm works in the compressed domain using H.264 encoded motion vectors, by which it saves on the resources spent to decode the video stream. The algorithm can be used to prune uninteresting frames, while the interesting frames can be send either to the cloud or processed further in the edge itself.

Keyphrases: IoT, pruning algorithm, surveillance

@booklet{EasyChair:6764,
  author    = {Lakshya and Venkata Suneel Kota and Mallikarjuna Rao Voleti and Shivraj Singh},
  title     = {Compressed Domain Consistent Motion Based Frame Scoring for IoT Edge Surveillance Videos},
  howpublished = {EasyChair Preprint 6764},
  year      = {EasyChair, 2021}}