# Robust Real-time Pedestrian Detection in Aerial Imagery on Jetson TX2

**Authors:** Mohamed Afifi, Yara Ali, Karim Amer, Mahmoud Shaker, Mohamed ElHelw  
**arXiv:** [1905.06653](https://arxiv.org/abs/1905.06653)  
**Date:** 2019-05-16  
**Summary page:** https://www.summarizepaper.com/en/paper/1905.06653v1/

## Key points

- Paper title: "Robust Real-time Pedestrian Detection in Aerial Imagery on Jetson TX2"
- Challenge: Detecting pedestrians in aerial imagery captured by drones
- Applications: Intersection monitoring, patrolling, surveillance
- Proposed framework: Pedestrian detection using YOLO object detection network
- High throughput: More than 5 frames per second (FPS) on Jetson TX2 embedded board
- Flexibility: Utilizes pre-trained model without additional training, easy application on different setups with minimal tuning
- Evaluation experiment: Applied to a sample video from the Embedded Real-Time Inference (ERTI) Challenge with unmanned aerial vehicle (UAV)
- Results: Average precision of approximately 81% mean Average Precision (mAP), accurately detects pedestrians in aerial imagery
- Comprehensive framework for real-time pedestrian detection in aerial imagery captured by drones
- Efficient and accurate detection using deep learning techniques and pre-trained models
- Potential applications: Intersection monitoring, surveillance

## Summary

The paper titled "Robust Real-time Pedestrian Detection in Aerial Imagery on Jetson TX2" addresses the challenge of detecting pedestrians in aerial imagery captured by drones. This task has various applications, including intersection monitoring, patrolling, and surveillance. To tackle these challenges, the authors propose a framework for pedestrian detection in videos using the YOLO (You Only Look Once) object detection network. The framework achieves a high throughput of more than 5 frames per second (FPS) on the Jetson TX2 embedded board, making it suitable for real-time applications. By leveraging deep learning techniques and utilizing a pre-trained model without requiring additional training, this flexibility allows for easy application on different setups with minimal tuning. In their experiments, the authors evaluate the proposed method by applying it to a sample video from the Embedded Real-Time Inference (ERTI) Challenge where pedestrians are monitored by an unmanned aerial vehicle (UAV). The method achieves an average precision of approximately 81% mean Average Precision (mAP), demonstrating its effectiveness in accurately detecting pedestrians in aerial imagery. Overall, this paper presents a comprehensive framework for real-time pedestrian detection in aerial imagery captured by drones. The use of deep learning techniques and pre-trained models enables efficient and accurate detection while running on resource-constrained embedded systems. The results obtained from applying this framework to real-world scenarios highlight its potential for various applications such as intersection monitoring and surveillance.