Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing the Fairness and Performance of Edge Cameras with Explainable AI

Truong Thanh Hung Nguyen, Vo Thanh Khang Nguyen, Quoc Hung Cao, Van Binh Truong, Quoc Khanh Nguyen, Hung Cao

TL;DR

Edge-camera AI for human detection often yields accurate but opaque models, hindering fairness and debugging. This work introduces an Explainable AI–driven debugging framework with expert-guided problem identification and solution design, validated on Bytetrack with YOLOX in real office edge networks. The approach reveals dataset bias toward full-body representations and occluded/disabled scenarios; after relabeling and bounding-box reannotation, the Bytetrack model shows improved localization and under-detection performance on challenging cases, including disabled individuals; the framework uses a structured, replicable workflow applicable to other detection tasks. Overall, the framework enables fairer, more trustworthy edge-detection systems and provides a practical pathway for diagnosing and improving complex detection models.

Abstract

The rising use of Artificial Intelligence (AI) in human detection on Edge camera systems has led to accurate but complex models, challenging to interpret and debug. Our research presents a diagnostic method using Explainable AI (XAI) for model debugging, with expert-driven problem identification and solution creation. Validated on the Bytetrack model in a real-world office Edge network, we found the training dataset as the main bias source and suggested model augmentation as a solution. Our approach helps identify model biases, essential for achieving fair and trustworthy models.

Enhancing the Fairness and Performance of Edge Cameras with Explainable AI

TL;DR

Edge-camera AI for human detection often yields accurate but opaque models, hindering fairness and debugging. This work introduces an Explainable AI–driven debugging framework with expert-guided problem identification and solution design, validated on Bytetrack with YOLOX in real office edge networks. The approach reveals dataset bias toward full-body representations and occluded/disabled scenarios; after relabeling and bounding-box reannotation, the Bytetrack model shows improved localization and under-detection performance on challenging cases, including disabled individuals; the framework uses a structured, replicable workflow applicable to other detection tasks. Overall, the framework enables fairer, more trustworthy edge-detection systems and provides a practical pathway for diagnosing and improving complex detection models.

Abstract

The rising use of Artificial Intelligence (AI) in human detection on Edge camera systems has led to accurate but complex models, challenging to interpret and debug. Our research presents a diagnostic method using Explainable AI (XAI) for model debugging, with expert-driven problem identification and solution creation. Validated on the Bytetrack model in a real-world office Edge network, we found the training dataset as the main bias source and suggested model augmentation as a solution. Our approach helps identify model biases, essential for achieving fair and trustworthy models.
Paper Structure (22 sections, 7 figures, 2 tables)

This paper contains 22 sections, 7 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Human Detection
  4. Explainable AI
  5. Debugging Model Framework with XAI
  6. Methodology
  7. Data Selection and Extraction of Predictions
  8. Statistical Analysis of Prediction Results
  9. Explanation Generation
  10. Problem Identification
  11. Solution Proposal
  12. Solution Assessment
  13. Model Enhancement
  14. Experiment
  15. Data Selection and Prediction Extraction
  16. ...and 7 more sections

Figures (7)

  • Figure 1: (a) A security camera on the ceiling of an office can detect ordinary people (green boxes), but not people who cover their bodies with a cloth. (b) The Bytetrack model cannot detect the disabled woman but still detect the other, who is not disabled.
  • Figure 2: The Debugging Framework for Human Detection Models
  • Figure 3: Examples of XAI Explanations with Bytetrack and YOLOX model. In which, each image in the second column is the XAI Explanations for a corresponding box.
  • Figure 4: Predictions of the Bytetrack model before and after fine-tuning.
  • Figure 5: Example of padding result. (Top, Left, Right, Bottom) = (100, 200, 200, 200) signifies padding of 100, 200, 200, and 200 pixels respectively on the top, left, right, and bottom.
  • ...and 2 more figures