# Aerial Images Processing for Car Detection using Convolutional Neural Networks: Comparison between Faster R-CNN and YoloV3

**Authors:** Adel Ammar, Anis Koubaa, Mohanned Ahmed, Abdulrahman Saad  
**arXiv:** [1910.07234](https://arxiv.org/abs/1910.07234)  
**Date:** 2019-10-16  
**Summary page:** https://www.summarizepaper.com/en/paper/1910.07234v1/

## Key points

- Authors address the challenge of car detection from aerial images using CNN
- Car detection from aerial images is more challenging than ground images
- Performance of two CNN algorithms, Faster R-CNN and YOLOv3, evaluated
- Impact of factors such as UAV's altitude, camera resolution, and object size on car detection accuracy assessed
- No algorithm outperforms the other in all cases
- Insights provided for selecting an appropriate algorithm based on dataset characteristics and application requirements
- Valuable insights into challenges and comprehensive evaluation of two CNN algorithms provided

## Summary

In the paper titled "Aerial Images Processing for Car Detection using Convolutional Neural Networks: Comparison between Faster R-CNN and YOLOv3," authors Adel Ammar, Anis Koubaa, Mohanned Ahmed, and Abdulrahman Saad address the challenge of car detection from aerial images using Convolutional Neural Networks (CNN). They highlight that car detection from aerial images is more challenging compared to ground images due to the difficulty in discerning vehicle features. To investigate this issue, the authors evaluate the performance of two state-of-the-art CNN algorithms: Faster R-CNN, a popular region-based algorithm; and YOLOv3, known for its fast detection capabilities. The authors analyze two datasets with different characteristics to assess the impact of various factors such as UAV's altitude, camera resolution, and object size on car detection accuracy. The objective of their work is to conduct a robust comparison between these cutting-edge algorithms. By utilizing a variety of metrics, they demonstrate that neither algorithm outperforms the other in all cases. This research contributes to advancing the field of car detection from aerial images by providing insights into the performance of two prominent CNN algorithms. The findings suggest that selecting an appropriate algorithm depends on specific factors such as dataset characteristics and application requirements. This information can guide researchers and practitioners in choosing an optimal approach for car detection from aerial imagery based on their specific needs. Overall, this paper provides valuable insights into the challenges associated with car detection from aerial images and offers a comprehensive evaluation of two state-of-the-art CNN algorithms. The results contribute to improving the accuracy and efficiency of car detection systems in aerial imagery applications.