Internship: Small Object Detection in Aerial Images

Le descriptif de l’offre ci-dessous est en Anglais_

**Niveau de diplôme exigé **:Bac + 4 ou équivalent

**Fonction **:Stagiaire de la recherche

**A propos du centre ou de la direction fonctionnelle**:
The Inria centre at Université Côte d'Azur includes 37 research teams and 8 support services. The centre's staff (about 500 people) is made up of scientists of different nationalities, engineers, technicians and administrative staff. The teams are mainly located on the university campuses of Sophia Antipolis and Nice as well as Montpellier, in close collaboration with research and higher education laboratories and establishments (Université Côte d'Azur, CNRS, INRAE, INSERM...), but also with the regiona economic players.

With a presence in the fields of computational neuroscience and biology, data science and modeling, software engineering and certification, as well as collaborative robotics, the Inria Centre at Université Côte d'Azur is a major player in terms of scientific excellence through its results and collaborations at both European and international levels.

**Contexte et atouts du poste**:
Knowledge of urban sewerage and drainage networks is essential for the rational mana
- gement of natural resources (pollution transfer), flood control and transportation network management. Acting as rainwater collectors, drainage networks can overflow at certain points and exacerbate flooding. However, most hydraulic models do not take into account the drai
- nage network, which is assumed to be rapidly saturated during a flood event. However, this is not due to a limitation of the models, but to a lack of knowledge of the characteristics of the networks. In fact, in many countries, whether developed or emerging, maps of networks or diagrams of structures are not widely available or are highly imperfect.

In France, local authorities are now responsible for wastewater treatment and must draw up a detailed description of wastewater and rainwater collection and transport facilities. But the maps available are still not very accurate, not very detailed and sometimes in analog format.

Image processing methods based on a geometric filter and a deep learning approach have demonstrated to be effective for detecting manhole covers on very high spatial resolution aerial images [2, 4, 5]. These works leveraged an early convolutional neural network model, AlexNet, to detect manholes from images at 5 cm spatial resolution acquired over two communes near Montpellier. The initial results showed that the manhole covers could be detected with an accuracy of 72% and a recall of 54% (which could be as high as 60%). Beyond the use of aerial imagery, several works have demonstrated that street-level imagery can be of great interest to detect and locate road objects [7]. The joint exploitation of aerial and street-level imagery has also been recently explored for enhancing aerial image segmentation [6], land use mapping for urban settlement [3] and traffic signs detection [1].

Several avenues of research could be explored to improve current manhole detection results, more precisely i) evaluate more recent deep learning frameworks for both semantic segmen
- tation and object detection and ii) integrate street-level imagery in a multi-modal analysis in order to assess the added value of this complementary information source for the process of manhole detection and location.

**Mission confiée**:
The aim of the internship will be to build upon our previous work on the detection of manhole covers through very high spatial resolution images using modern deep neural networks.

**Principales activités**:
The aim of the internship will be to build upon our previous work on the detection of manhole covers through very high spatial resolution images using modern deep neural networks. More specifically, it will involve:

- — Train and validate a network on the images available (and if possible, on other images of lower resolution, available in open source).
- — Reduce as much as possible possible false positives by verifying the actual presence of manhole covers detected on street-view images.

**Compétences**:

- — Master 2 in computer science or signal processing, with good knowledge on machine learning and/or image analysis.
- — Tools and programming language : Python (preferably) and familiarity with some deep learning library (PyTorch, Tensorflow).

**Avantages**:

- Subsidized meals
- Partial reimbursement of public transport costs
- Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
- Possibility of teleworking (after 6 months of employment) and flexible organization of working hours
- Professional equipment available (videoconferencing, loan of computer equipment, etc.)
- Social, cultural and sports events and activities
- Access to vocational training
- Social security coverage