Quay phim trên không với các drone đồng bộ hóa sử dụng học tăng cường

Multimedia Tools and Applications - Tập 80 - Trang 18125-18150 - 2021
Kenneth C. W Goh1, Raymond B. C Ng1, Yoke-Keong Wong1, Nicholas J. H Ho1, Matthew C. H Chua1
1Institute of Systems Science, National University of Singapore, Singapore, Singapore

Tóm tắt

Việc sử dụng nhiều drone là cần thiết cho các ứng dụng quay phim trên không để đảm bảo tính dự phòng. Tuy nhiên, điều này có thể làm tăng nguy cơ va chạm, đặc biệt là khi số lượng drone tăng lên. Do đó, điều này thúc đẩy chúng tôi khám phá các phương pháp kiểm soát hình thức bay tự động khác nhau có tiềm năng cho phép nhiều drone theo dõi một mục tiêu cụ thể một cách hiệu quả cùng một lúc. Trong bài báo này, chúng tôi đã thiết kế một thuật toán học tăng cường sâu không cần mô hình, chủ yếu dựa trên khái niệm Mạng Q Hồi tiếp Sâu, cho các mục đích đã nêu. Thuật toán đề xuất được mở rộng thành các loại tác nhân đơn và đa tác nhân cho phép nhiều drone theo dõi trong khi duy trì hình thức và ngăn ngừa va chạm. Các phần thưởng liên quan trong các phương pháp này có bản chất hai chiều và phụ thuộc vào hệ thống truyền thông. Sử dụng bộ mô phỏng Microsoft AirSim, một môi trường ảo bao gồm bốn drone ảo đã được phát triển cho các mục đích thử nghiệm. Một so sánh đã được thực hiện giữa các phương pháp khác nhau trong quá trình mô phỏng, và kết quả cho thấy rằng mô hình tác nhân đơn hồi tiếp là phương pháp hiệu quả nhất, hiệu quả hơn 33% so với các phiên bản đa tác nhân hồi tiếp của nó. Hiệu suất kém của mô hình tác nhân đơn không hồi tiếp cũng gợi ý rằng các yếu tố hồi tiếp trong mạng là rất cần thiết để cho phép hoạt động bay của nhiều drone theo cách mong muốn.

Từ khóa

#drone #quay phim trên không #học tăng cường sâu #kiểm soát hình thức bay #mô phỏng

Tài liệu tham khảo

Abadi M, Agarwal A, Barham P et al (2016) TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems. Cornell University. arXiv:1603.04467. Accessed 1 Jun 2020 Abughalieh KM, Sababha BH, Rawashdeh NA (2019) A video-based object detection and tracking system for weight sensitive UAVs. Multimed Tools Appl 78:9149–9167. https://doi.org/10.1007/s11042-018-6508-1 Alam MS, Natesha BV, Ashwin TS, et al. (2019) UAV Based cost-effective real-time abnormal event detection using edge computing. Multimed Tools Appl 78:35119–35134. https://doi.org/10.1007/s11042-019-08067-1 Becker-Ehmck P, Karl M, Peters J et al (2020) Learning to Fly via Deep Model-Based Reinforcement Learning. Cornell University. arXiv:2003.08876. Accessed 1 Jun 2020 Bochkovskiy A, Wang CY, Liao HYM (2020) YOLOv4: Optimal Speed and Accuracy of Object Detection. Cornell University. arXiv:2004.10934. Accessed 1 Jun 2020 Bonatti R, Ho C, Wang W et al (2019) Towards a Robust Aerial Cinematography Platform: Localizing and Tracking Moving Targets in Unstructured Environments. Cornell University. arXiv:1904.02319. Accessed 1 Jun 2020 Bonatti R, Zhang Y, Choudhury S et al (2018) Autonomous drone cinematographer: Using artistic principles to create smooth, safe, occlusion-free trajectories for aerial filming. Cornell University. arXiv:1808.09563. Accessed 1 Jun 2020 Chollet F et al (2019) Keras. https://keras.io. Accessed 1 Jun 2020 Cunha R (2017) ICCV2017 Tutorial: Drone vision for cinematography - Drone Formation and Flight Control. MultiDrone. https://multidrone.eu/wp-content/uploads/2017/01/04_Drone_Formation_Flight_Control.pdf. Accessed 1 Jun 2020 Epic Games (2020) Unreal Engine. https://www.unrealengine.com. Accessed 1 Jun 2020 Esmukov K, Tygart A, López A et al (2018) Geocoding library for Python. GitHub repository. https://github.com/geopy/geopy. Accessed 1 Jun 2020 French S (2018) Want to make six figures? Try being a drone pilot. MarketWatch. https://www.marketwatch.com/story/want-to-make-six-figures-try-being-a-drone-pilot-2018-08-10. Accessed 1 Jun 2020 Galvane Q, Fleureau J, Tariolle FL, Guillotel P (2017) Automated Cinematography with Unmanned Aerial Vehicles. Cornell University. arXiv:1712.04353. Accessed 1 Jun 2020 Galvane Q, Lino C, Christie M et al (2018) Directing cinematographic drones. ACM Trans Graph 37(3):1–18. https://doi.org/10.1145/3181975 Hausknecht M, Stone P (2017) Deep recurrent Q-Learning for partially observable MDPs. Cornell University. arXiv:1507.06527. Accessed 1 Jun 2020 Hong S (2019) Autonomous UAV Navigation without Collision using Visual Information in Airsim. GitHub repository. https://github.com/sunghoonhong/AirsimDRL. Accessed 1 Jun 2020 Huang C, Gao F, Pan J, et al. (2018) ACT: An autonomous drone cinematography system for action scenes. 2018 IEEE international conference on robotics and automation (ICRA). Brisbane 7039–7046 https://doi.org/10.1109/ICRA.2018.8460703 Huang C, Yang Z, Kong Y et al (2018) Through-the-lens Drone Filming. 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Madrid 4692–4699 https://doi.org/10.1109/IROS.2018.8594333 Huynh NA (2017) Training and Detecting Objects with YOLO3. GitHub repository. https://github.com/experiencor/keras-yolo3. Accessed 1 Jun 2020 James S, Freese M, Davison AJ (2019) PyRep: Bringing V-REP to Deep Robot Learning. Cornell University. arXiv:1906.11176. Accessed 1 Jun 2020 James S, Ma Z, Arrojo DR, Davison AJ (2019) RLBench: The Robot Learning Benchmark & Learning Environment. Cornell University. arXiv:1909.12271. Accessed 1 Jun 2020 Joubert N, Jane LE, Goldman DB, Berthouzoz F et al (2016) Towards a Drone Cinematographer: Guiding Quadrotor Cameras using Visual Composition Principles. Cornell University. arXiv:1610.01691. Accessed 1 Jun 2020 Karney CFF (2013) Algorithms for geodesics. J Geodesy 87:43–55. https://doi.org/10.1007/s00190-012-0578-z Kostrikov I, Yarats D, Fergus R (2020) Image Augmentation Is All You Need: Regularizing Deep Reinforcement Learning from Pixels. Cornell University. arXiv:2004.13649. Accessed 1 Jun 2020 Krishnan S, Boroujerdian B, Fu W et al (2019) Air Learning: An AI Research Platform for Algorithm-Hardware Benchmarking of Autonomous Aerial Robots. Cornell University. arXiv:1906.00421. Accessed 1 Jun 2020 Kwak J, Park JH, Sung Y (2019) Affective social big data generation algorithm for autonomous controls by CRNN-based end-to-end controls. Multimed Tools Appl 78:27175–27192. https://doi.org/10.1007/s11042-019-7703-4 Liu F, Yang A (2019) Application of gcForest to visual tracking using UAV image sequences. Multimed Tools Appl 78:27933–27956. https://doi.org/10.1007/s11042-019-07864-y Liu H, Zhao W, Lewis L et al (2019) Attitude Synchronization for Multiple Quadrotors using Reinforcement Learning*. 2019 Chinese Control Conference (CCC). Guangzhou 2480–2483 https://doi.org/10.23919/ChiCC.2019.8865177 Mademlis I, Nikolaidis N, Tefas A, et al. (2018) Autonomous unmanned aerial vehicles filming in dynamic unstructured outdoor environments. IEEE Signal Proc Mag 36(1):147–153. https://doi.org/10.1109/MSP.2018.2875190 Microsoft (2019) Microsoft Drone Rescue. GitHub repository. https://github.com/microsoft/DroneRescue. Accessed 1 Jun 2020 Microsoft (2020) Open source simulator for autonomous vehicles built on Unreal Engine / Unity, from Microsoft AI & Research. GitHub repository. https://github.com/microsoft/AirSim. Accessed 1 Jun 2020 Nägeli T, Alonso-Mora J, Domahidi A et al (2017) Real-Time Motion planning for aerial videography with dynamic obstacle avoidance and viewpoint optimization. IEEE Robot Autom Lett 2(3):1696–1703. https://doi.org/10.1109/LRA.2017.2665693 Nägeli T, Meier L, Domahidi A, et al. (2017) Real-time planning for automated multi-view drone cinematography. ACM Trans Graph 36 (4):1–10. https://doi.org/10.1145/3072959.3073712 Nikolaidis N, Mademlis I, Raptopoulou C, Bull D (2017) ICCV2017 Tutorial: Drone vision for cinematography - Drone Cinematography. MultiDrone. https://multidrone.eu/wp-content/uploads/2017/01/06_Drone-Cinematography.pdf. Accessed 1 Jun 2020 Passalis N, Tefas A (2019) Deep reinforcement learning for controlling frontal person close-up shooting. Neurocomputing 335:37–47. https://doi.org/10.1016/j.neucom.2019.01.046 Redmon J, Farhadi A (2018) YOLOv3: An Incremental Improvement. Cornell University. arXiv:1804.02767. Accessed 1 Jun 2020 Rising J (2015) Drones vs Helicopter - Part 1: How drones are changing the aerial video industry. Flight Evolved. https://flight-evolved.com/drone-vs-helicopter. Accessed 1 Jun 2020 Sabetghadam B, Alcántara A, Capitán J et al (2019) Optimal trajectory planning for autonomous drone cinematography. 2019 european conference on mobile robots (ECMR). Prague 1–7 https://doi.org/10.1109/ECMR.2019.8870950 Tefas A, Nousi P, Passalis N et al (2017) ICCV2017 Tutorial: Drone vision for cinematography - Deep Learning for Drone Vision in Cinematography. MultiDrone. https://multidrone.eu/wp-content/uploads/2017/01/05_Deep-Learning-for-Drone-Vision-in-Cinematography.pdf. Accessed 1 Jun 2020 Torres-González A, Capitán J, Cunha R et al (2017) A multidrone approach for autonomous cinematography planning. In: In: Ollero A, Sanfeliu A, Montano L et al (eds) ROBOT 2017: Third iberian robotics conference. ROBOT 2017. Advances in intelligent systems and computing, vol 693. Springer, Cham, pp 337–349, DOI https://doi.org/10.1007/978-3-319-70833-1_28 Tzutalin (2020) LabelImg is a graphical image annotation tool and label object bounding boxes in images. GitHub repository. https://github.com/tzutalin/labelImg. Accessed 1 Jun 2020 Wang T, Qin R, Chen Y et al (2019) A reinforcement learning approach for UAV target searching and tracking. Multimed Tools Appl 78:4347–4364. https://doi.org/10.1007/s11042-018-5739-5 Yang H, Xie K, Huang S, Huang H (2018) Uncut aerial video via a single sketch. Comput Graph Forum 37:191–199. https://doi.org/10.1111/cgf.13559 Zanol R, Chiariotti F, Zanella A (2019) Drone mapping through multi-agent reinforcement learning. 2019 IEEE Wireless Communications and Networking Conference (WCNC). Marrakesh 1–7 https://doi.org/10.1109/WCNC.2019.8885873 Zhang K, Yang Z, Basar T (2019) Multi-Agent Reinforcement Learning: A Selective Overview of Theories and Algorithms. Cornell University. arXiv:1911.10635. Accessed 1 Jun 2020