Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Real-Time AI-Driven People Tracking and Counting Using Overhead Cameras

Ishrath Ahamed, Chamith Dilshan Ranathunga, Dinuka Sandun Udayantha, Benny Kai Kiat Ng, Chau Yuen

TL;DR

This study proposes a novel approach combining a new object tracking algorithm, a novel counting algorithm, and a fine-tuned object detection model that achieves 97% accuracy in real-time people counting with a frame rate of 20–27 FPS on a low-power edge computer.

Abstract

Accurate people counting in smart buildings and intelligent transportation systems is crucial for energy management, safety protocols, and resource allocation. This is especially critical during emergencies, where precise occupant counts are vital for safe evacuation. Existing methods struggle with large crowds, often losing accuracy with even a few additional people. To address this limitation, this study proposes a novel approach combining a new object tracking algorithm, a novel counting algorithm, and a fine-tuned object detection model. This method achieves 97% accuracy in real-time people counting with a frame rate of 20-27 FPS on a low-power edge computer.

Real-Time AI-Driven People Tracking and Counting Using Overhead Cameras

TL;DR

This study proposes a novel approach combining a new object tracking algorithm, a novel counting algorithm, and a fine-tuned object detection model that achieves 97% accuracy in real-time people counting with a frame rate of 20–27 FPS on a low-power edge computer.

Abstract

Accurate people counting in smart buildings and intelligent transportation systems is crucial for energy management, safety protocols, and resource allocation. This is especially critical during emergencies, where precise occupant counts are vital for safe evacuation. Existing methods struggle with large crowds, often losing accuracy with even a few additional people. To address this limitation, this study proposes a novel approach combining a new object tracking algorithm, a novel counting algorithm, and a fine-tuned object detection model. This method achieves 97% accuracy in real-time people counting with a frame rate of 20-27 FPS on a low-power edge computer.

Paper Structure

This paper contains 8 sections, 2 equations, 3 figures, 3 tables, 1 algorithm.

Table of Contents

  1. INTRODUCTION
  2. METHODOLOGY
  3. Head Detection
  4. Feature Extraction
  5. Tracking
  6. Counting Algorithm
  7. RESULTS AND DISCUSSION
  8. CONCLUSION

Figures (3)

  • Figure 1: Before specifically training our model to identify potential disturbances, there were few mismatches, identifying distractions as potential head objects. For example in (a) the chair is identified as a person with 95%. But after training (b) the chair is identified as a chair with 89% confidence, while the person is identified with 97% & 94% confidence in both cases respectively. (c) people detection in low/no light conditions. The model correctly detects the two people in the IR camera frame with 70% (lower) and 58% (top) confidence.
  • Figure 2: Object Tracking Pipeline
  • Figure 3: Real time deployment result can be found here real_time_results: (a) before entering the room. "# of Ins" is 9. (b) After two people, the man wearing a red T-shirt and the man wearing a green T-shirt, entering the room, "# of Ins" has been increased by 2 indicating 2 people entered the room. Also their assigned IDs change from A to C indicating the shift from outside to inside.