Published:  6th, January, 2026

Beyond the Gaze: Automated Detection of Potential Areas of Interest On Visual Stimuli

Fatima Isiaka,  Samuel Esther Alu,  AHMED, Mustafa Umar  and  Muhammad Abdullahi Yau

Abstract

From the static images on our screens to the dynamic videos that stream through our devices, a fundamental question persists on what captures our attention. Understanding where people look and for how long is the cornerstone of effective design and even cognitive research. Traditionally, determining ”Areas of Interest” (AOIs) has relied on arduous methods like manual annotation or expensive, often intrusive, eye-tracking studies. But what if we could predict, with increasing accuracy and speed, where potential AOIs lie, before a single human eye even lands on the stimulus? This paper introduces a novel approach, the Randomised Object Detection Algorithm (RODA), that leverages randomisation to explore the visual landscape in a less biased way, rather than relying on a fixed set of parameters or a single, deterministic approach. The result shows that RODA achieved an average AOI localisation accuracy increase of 18% compared to standard contrast-based saliency maps; its performance dropped in fixed images of weed detection rice field, with  only 5%, compared to drops up to an aggregate of all like images with 25% from conventional models and was very effective in generalising object detection.

Keywords: Randomization, Object Detection, Area of Interest, Saliency Mapping, Ingredients of Attention, Weed Detection