A Convolutional Neural Network for Real-Time Vehicle Detection Under the Unmanned Aerial Vehicle Platform
Shikun Chen, Xiaobo Lu, Yongbin Li, Renliang Wu · 2019
The detection of vehicles under the unmanned aerial vehicle platform has several technical challenges. Compared with surveillance videos, aerial videos have more complex background and broader range which lead to larger space for searching. It is also difficult to build a effective background model due to the movement of the unmanned aerial vehicle. The low performance of development board on the aerial vehicle also make it difficult to perform real-time detection. We proposes a convolutional neural network to carry out real-time detection of vehicles under the unmanned aerial vehicle platform. Multi-scale anchor design is carried out to improve the adaptability to different vehicles and the algorithm is time optimized based on the binary weight network. As a consequence, our algorithm runs at 28.8 FPS with average precision of 91.2% under the unmanned aerial vehicle platform.