Critiquing AI Surveys with a Case Study on Temporal Features for Trustworthy Computer Vision

Abstract

A novel, thorough manual survey is presented, focusing on the trustworthy extraction and application of temporal features in drone video feeds for object detection and tracking. Trust stems from reducing the necessary scope of opaque, deep components, and externally accounting for hallucinations. This manual survey is used as a baseline for a first of its kind case study into the use of new end-to-end LLM based “Deep Research” tools for literature review purposes. It was found that such tools cannot currently be used to replace a manual survey on novel research topics, but may offer utility gauging method popularity.