ORGANISATION/COMPANYUniversité Gustave Eiffel
RESEARCH FIELDComputer scienceMathematics › Applied mathematics
RESEARCHER PROFILEFirst Stage Researcher (R1)
APPLICATION DEADLINE21/03/2022 17:00 - Europe/Brussels
LOCATIONFrance › Marne-La-Vallée
TYPE OF CONTRACTTemporary
HOURS PER WEEK35
OFFER STARTING DATE01/10/2022
EU RESEARCH FRAMEWORK PROGRAMMEH2020 / Marie Skłodowska-Curie Actions
MARIE CURIE GRANT AGREEMENT NUMBER101034248
One of the core tasks of the takeaway business is to deliver goods to customers in a timely and non-destructive manner. However, with the continuous increase in the number of takeaway orders, traffic congestion, and shortage of delivery personnel in big cities have become increasingly prominent. It has become a big challenge to deliver goods to customers in a timely manner. In order to solve the above problems, the demand for drone delivery has emerged. Unmanned Aerial Vehicle (UAV) delivery has the characteristics of zero congestion, zero contact, and short paths. With appropriate infrastructure, it can even complete tasks such as delivery to communities and direct delivery to households. As early as 2013, Amazon founder Bezos predicted that drone delivery will become the mainstream of business in the future, while Uber and United Parcel Service (UPS), and FedEx have been rapidly advancing drone commercial delivery projects, towards the development of smart cities.
Currently, drone delivery has not reached the expected development speed. The reasons include the complexity of the scene, the insufficient safety guarantee performance, the excessive weight of accessories, etc.. More and more effort has been devoted to developing fast and accurate visual analysis algorithms that could guide the flight of UAVs. However, the state-of-the-art performance is still not good enough for real applications, mainly due to the challenges of visual perception.
In visual perception, scene understanding of images/videos captured by a static camera is one core research topic, and has a large amount of existing works. The study of visual perception for autonomous driving (AD) has also drawn more and more attention, where the quick movement of the car and its camera(s) significantly increases the difficulty of scene understanding. Visual perception for UAVs is even more challenging in general, mainly due to: (1) The 3D motion of drones leads to much more dynamic, complex, and flexible scenes; (2) On-board processing is safer than remote processing, due to the instability of the network signal. However, for common drones, the computing capacity is far less than that of unmanned cars, due to the constraints in size, weight capacity, etc.. This imposes a requirement on the efficiency of visual perception algorithms.
In the aforementioned context, as a long-term project, we aim to study visual perception in the context of UAVs, and investigate a series of crucial problem (e.g., SLAM, scene reconstruction, scene recognition, semantic segmentation, autopilot, etc.) for substantially boosting the development of UAVs, towards their applications in smart cities.
In this thesis, we focus on the scenario of UAV safe landing in urban areas (e.g., [1,2]), where visual perception is more crucial and challenging than other scenarios (e.g., high-altitude flying, landing in open areas, etc.), due to the high complexity of the scene, including static objects of various shapes (e.g., buildings, steady trees, etc.), moving objects (e.g., cars, pedestrians, animals, etc.), limited space for flying, etc.. The research goal is to study two representative problems and boost their states of the art:
(1) Depth map estimation . The depth map provides each pixel of the image with the vertical distance from the three-dimensional point corresponding to the pixel to the center of the camera capturing the image, which is an essential feature of the three-dimensional world exhibited in the two-dimensional image.
(2) Camera motion estimation . This provides a highly valuable feature to estimate UAV’s orientation.
Moreover, the estimation of both the depth map and the camera motion offers advanced visual information for UAV to understand the three-dimensional world around it, and estimate the relative pose w.r.t. the target landing place, which is essential for safe and precise landing (including related tasks such as subsequent path planning, obstacle avoidance, etc.).
Below are three steps towards the goal:
(1) Study of efficient and advanced methods for the two problems. In particular, we will investigate how to efficiently and effectively exploit the complementarity between the two problems so as to jointly estimate the depth map and the camera motion together in an efficient manner. In specific, we plan to explore the temporal relations between adjacent frames to infer the depth map and camera motion jointly in a Bayesian framework. Considering the complexity of scenes, we will model occlusion, moving objects, and illumination in a unified framework.
(2) Investigation of efficient domain adaptation methods  to allow fast and better generalization of the learned model to new scenarios (e.g, new cities, different seasons, unseen weather conditions, etc.), since the training dataset is only a sparse subset of the whole data space.
(3) Development of computational and memory-efficient light-weight networks for onboard inference of depth and camera motion. To this end, we will explore network binarization and compression techniques to balance the efficiency and the estimation accuracy. In addition, we will exploit knowledge distillation to improve the estimation accuracy of light-weight networks by transferring knowledge from large networks.
The research results will provide insights to address other UAV visual perception problems and the aforementioned less challenging contexts, and also lay a foundation for us to work on more and more UAV-related projects, toward the development of smart cities.
 Castellano, Giovanna, Ciro Castiello, Corrado Mencar, and Gennaro Vessio. ""Crowd detection for drone safe landing through fully-convolutional neural networks."" In International conference on current trends in theory and practice of informatics, 2020.
 Gonzalez-Trejo, J.A. and Mercado-Ravell, D.A. “Lightweight density map architecture for uavs safe landing in crowded areas.” Journal of Intelligent & Robotic Systems, 2021.
 Fu, Huan, Mingming Gong, Chaohui Wang, Kayhan Batmanghelich, and Dacheng Tao. ""Deep ordinal regression network for monocular depth estimation."" In IEEE Conference on Computer Vision and Pattern Recognition, 2018.
 Tang, Jiexiong, John Folkesson, and Patric Jensfelt. ""Geometric correspondence network for camera motion estimation."" IEEE Robotics and Automation Letters, 2018.
 Fu, Huan, Mingming Gong, Chaohui Wang, Kayhan Batmanghelich, Kun Zhang, and Dacheng Tao. "
A secondment at Melbourne Centre for Data Science, The University of Melbourne, Australia.
For more information, contact the PhD thesis supervisor.
- High-quality doctoral training rewarded by a PhD degree, delivered by Université Gustave Eiffel
- Access to cutting-edge infrastructures for research & innovation.
- Appointment for a period of 36 months based on a salary of 2 700 € (gross salary per month).
- Job contract under the French labour legislation in force, respecting health and safety, and social security: 35 hours per week contract, 25 days of annual leave per year.
- International mobility will be mandatory
- An international environment supported by the adherence to the European Charter & Code.
- Access to dedicated CLEAR-Doc trainings with a strong interdisciplinary focus, together with a Career development Plan.
- At the time of the deadline, applicants must be in possession or finalizing their Master’s degree or equivalent/postgraduate degree. At the time of recruitment, applicants must be in possession of their Master’s degree or equivalent/postgraduate degree which would formally entitle to embark on a doctorate.
- At the time of the deadline, applicants must be in the first four years (full-time equivalent research experience) of their research career (career breaks excluded) and not yet been awarded a doctoral degree. Career breaks refer to periods of time where the candidate was not active in research, regardless of his/her employment status (sick leave, maternity leave etc). Short stays such as holidays and/or compulsory national service are not taken into account.
- At the time of the deadline, applicants must not have resided or carried out their main activity (work, studies, etc.) in France for more than 12 months in the 3 years immediately prior to the call deadline.
- Applicants must be available to start the programme on schedule (around 1st October 2022).
- Please refer to the Guide for Applicants available on the CLEAR-Doc website.
- The First step before applying is contacting the PhD supervisor. You will not be able to apply without an acceptation letter from the PhD supervisor.
- Please contact the PhD supervisor for any additional detail on job offer.
- There are no restrictions concerning the age, gender or nationality of the candidates. Applicants with career breaks or variations in the chronological sequence of their career, with mobility experience or with interdisciplinary background or private sector experience are welcome to apply.
- Support service is available during every step of the application process by email: firstname.lastname@example.org
Web site for additional job details
REQUIRED EDUCATION LEVELComputer science: Master Degree or equivalent
REQUIRED LANGUAGESENGLISH: Excellent
- At the time of the deadline, applicants must be in possession or finalizing their Master’s degree or equivalent/postgraduate degree.
- At the time of recruitment, applicants must be in possession of their Master’s degree or equivalent/postgraduate degree which would formally entitle to embark on a doctorate.
EURAXESS offer ID: 717584
The responsibility for the jobs published on this website, including the job description, lies entirely with the publishing institutions. The application is handled uniquely by the employer, who is also fully responsible for the recruitment and selection processes.
Please contact email@example.com if you wish to download all jobs in XML.