In this article, we present an educational reform we implemented a few years ago to respond to a marked drop in the success of students entering higher education in computer science. The main objective of our reform is to adapt the teaching methods of Generation X-Y to Generation Y-Z or more. To do this, we propose two approaches. First, to make learning more active, stimulating and empowering. Second, to make learning more individualized in a context of a large group of students with an optimization of the teacher’s time. We first present an analysis of the probable reasons for the lower level of students and the specific issues encountered by both students and teachers. Then, we detail how we implemented these solutions in the form of an original e-learning platform based on two back-end tools able to manage large number of students: an efficient real-time auto-corrector of source codes and a very robust anti-plagiarism software based on computer distance information theory. Finally, we present the results obtained after an experimentation of 8 years. These results are put into perspective by an analysis of 15 years of qualitative and quantitative indicators carried out on hundreds of students each year. The analysis shows that the technical skills and the involvement of students are improved even in groups of several hundreds of students. Computer science teachers wishing to quickly and concretely develop the involvement of their students can rely on these tried and tested key levers and solutions presented.

Tasks such as distinguishing or identifying individual objects of interest require the production of dense local clouds at the scale of these individual objects of interest. Due to the physical and dynamic properties of an underwater environment, the usual dense matching algorithms must be rethought in order to be adaptive. These properties also imply that the scene must be observed at close range. Classic robotized acquisition systems are oversized for local studies in shallow water while the systematic acquisition of data is not guaranteed with divers. We address these two major issues through a multidisciplinary approach. To efficiently acquire on-demand stereoscopic pairs using simple logistics in small areas of shallow water, we devised an agile light-weight dedicated system which is easy to reproduce. To densely match two views in a reliable way, we devised a reconstruction algorithm that automatically accounts for the dynamics, variability and light absorption of the underwater environment. Field experiments in the Mediterranean Sea were used to assess the results.

Assessing the completeness of an underwater 3D reconstruction on-site is crucial as it allows for rescheduling acquisitions, which capture missing data during a mission, avoiding additional costs of a subsequent mission. This assessment needs to rely on a dense point cloud since a sparse cloud lacks detail and a triangulated model can hide gaps. The challenge is to generate a dense cloud with field-deployable tools. Traditional dense reconstruction methods can take several dozen hours on low-capacity systems like laptops or embedded units. To speed up this process, we propose building the dense cloud incrementally within an SfM framework while incorporating data redundancy management to eliminate recalculations and filtering already-processed data. The method evaluates overlap area limits and computes depths by propagating the matching around SeaPoints—the keypoints we design for identifying reliable areas regardless of the quality of the processed underwater images. This produces local partial dense clouds, which are aggregated into a common frame via the SfM pipeline to produce the global dense cloud. Compared to the production of complete dense local clouds, this approach reduces the computation time by about 70% while maintaining a comparable final density. The underlying prospect of this work is to enable real-time completeness estimation directly on board, allowing for the dynamic re-planning of the acquisition trajectory.

The 3D reconstruction of underwater scenes from overlapping images requires modeling the sensor. While underwater self-calibration gives good results when coupled with multi-view algorithms, calibration or pre-calibration with a pattern is still necessary when scenes are weakly textured or if there are not enough points of view of the same points; however, detecting patterns on underwater images or obtaining a good distribution of these patterns on a dataset is not an easy task. Thus, we propose a methodology to guide the acquisition of a relevant underwater calibration dataset. This process is intended to provide feedback in near real-time to the operator to guide the acquisition and stop it when a sufficient number of relevant calibration images have been reached. To perform this, pattern detection must be optimized both in time and success rate. We propose three variations of optimized detection algorithms, each of which takes into account different hardware capabilities. We present the results obtained on a homemade database composed of 60,000 images taken both in pools and at sea.

Close-range remote sensing, and more particularly, its acquisition part that is linked to field robotics, is at the crossroads of many scientific and engineering fields. Thus, it takes time for students to acquire the solid foundations needed before practicing on real systems. Therefore, we are interested in a means that allow students without prerequisites to quickly appropriate the fundamentals of this interdisciplinary field. For this, we adapted a haggle game to the close-range remote sensing theme. In this article, we explain the mechanics that serve our educational purposes. We have used it, so far, for four academic years with hundreds of students. The experience was assessed through quality surveys and quizzes to calculate success indicators. The results show that the serious game is well appreciated by the students. It allows them to better structure information and acquire a good global vision of multi-domain acquisition and data processing in close-range remote sensing. The students are also more involved in the rest of the lessons; all of this helps to facilitate their learning of the theoretical parts. Thus, we were able to shorten the time before moving on to real practice by replacing three lesson sessions with one serious game session, with an increase in mastering fundamental skills. The designed serious game can be useful for close-range remote sensing teachers looking for an effective starting lesson. In addition, teachers from other technical fields can draw inspiration from the creation mechanisms described in this article to create their own adapted version. Such a serious game is also a good asset for selecting promising students in a recruitment context.