04-17-2021(SUN) ~ 08-05-2021(FRI)

Recently, as the demand for delivery has increased, there has been a lot of discussions about services that effectively deliver to users. In this situation, research in drone delivery service is improved. In the case of drones, they are more environmentally friendly than other delivery methods. Moreover, drone delivery systems can figure out the wrong delivery problem the reason why algorithms operate them and there is no traffic jam. Hence, consumers can receive delivery services quickly. However, if the demand for drone delivery services would increase, much discussions will be needed about drone traffic. The biggest problem here is when a collision occurs between the drones being delivered. Since accidents between drones occur in the air, they can bring a relatively greater risk than accidents on land.

• Software : Develop an algorithm that can estimate the position of UAV for sensing and avoiding in drone delivery system.
• Hardware : Using camera and microphone which can detect and estimate the target drone cost-efficiently.

According to previous studies, conventional drone detection technologies such as Radar or Lidar are expensive, so cost-effective vision-based detection methods are mainly used comparing to the traditional method. However, in the case of vision, it is difficult to distinguish between flying objects such as birds and drones. Hence, drone detection is possible more accurately by using the microphone as an additional detection method to check if the drone approaches.

Most previous research in drone detection field only focused on single domain. Not only research in the drone detection task, but also studies in position estimation of drone are only interested in single domain. However, the studies about multi-modality are increased and be one of the hot topic. In this point of view, individual domains and multi-domain could be conducted

1. The visual and acoustic data are collected via camera and microphone.
2. The pre-processing is conducted on visual and acoustic data.
3. The pro-processed data is labeled with distance from detecting drone.
4. Deep learning based method would be used for training model.
5. The combination method using visual and acoustic method could be compared with only vision or acoustic method.
6. It will be deployed in drone delivery service via API.

• Google Colaboratory
• Python version 3.10
• OpenCV CV2 2.4
• MacBook Air 2021 M1
• DJI MAVIC 2 Pro
• EVO 2

• 👩‍💻 Dongwhan Lee

  • Kyunghee Univeristy
  • Major in Software Convergence
  • derick_lee@khu.ac.kr

• 👰🏻 Yeongseo Kim

  • Sangmyung University
  • Major in Human Centered-Artificial Intelligence
  • 201910787@sangmyung.kr

• 👨🏻‍💻 Juann Kim

  • Sangmyung University
  • Major in Department of Software
  • 201920951@sangmyung.kr

• 👩‍🚀 Heeyeon Shin

  • Kyunghee University
  • Major in Computer Science Engineering and International Studies
  • 567didi@khu.ac.kr

• 👩‍🚀 Yeeun Heo

  • Soongsil University
  • Major in Software Engineering
  • gjdpdms2005@soongsil.ac.kr

• 👨🏻‍ Luke Siemers

  • Purdue Univeristy
  • Major in Computer and Information Technology
  • lukesiemers33@gmail.com