Author(s):
1. HOANH NGUYEN:
Faculty of Electrical Engineering Technology, Industrial University of Ho Chi Minh City,Ho Chi Minh, City, Vietnam
Abstract:
Although deep learning-based object detectors have achieved great success in general object detection in recent years, detecting of objects like car in aerial images is still a challenge. The main difficulty of car detection in aerial images comes from the relatively small size with multiple orientations of car in images. In addition, due to the high resolution of aerial images, the inference time of current approaches is still high. To solve these problems, this paper proposes an enhanced framework for fast and efficient car detection in aerial images. In the proposed approach, ResNet-34 architecture is adopted to create the base convolution layers. Compared with ResNet-50 and ResNet-101, ResNet-34 achieves comparable performance while being faster and simple. Then, an enhanced feature map generation module is designed to generate enhanced feature maps from input feature maps. To speed up the detection process, the detection network based on region proposal network is used to exactly locate cars in original aerial images. The detection network included region proposal networks is applied at different enhanced feature maps with different scales to detect multi-scale car in input image. Experimental results on public dataset show that the proposed approach achieves comparable performance compared with other state-of-the-art approaches.
Page(s):
889-899
DOI:
DOI not available
Published:
Journal: Journal of Theoretical and Applied Information Technology, Volume: 98, Issue: 5, Year: 2020
Keywords:
Convolutional Neural Network
,
Car Detection
,
Object Detection
,
Pyramid Network
,
Intelligent Transportation System
References:
References are not available for this document.
Citations
Citations are not available for this document.