Sébastien Piérard

Characterizing domains is essential for models analyzing dynamic environments, as it allows them to adapt to evolving conditions or to hand the task over to backup systems when facing conditions outside their operational domain. Existing solutions typically characterize a domain by solving a regression or classification problem, which limits their applicability as they only provide a limited summarized description of the domain. In this paper, we present a novel approach to domain characterization by characterizing domains as probability distributions. Particularly, we develop a method to predict the likelihood of different weather conditions from images captured by vehicle-mounted cameras by estimating distributions of physical parameters using normalizing flows. To validate our proposed approach, we conduct experiments within the context of autonomous vehicles, focusing on predicting the distribution of weather parameters to characterize the operational domain. This domain is characterized by physical parameters (absolute characterization) and arbitrarily predefined domains (relative characterization). Finally, we evaluate whether a system can safely operate in a target domain by comparing it to multiple source domains where safety has already been established. This approach holds significant potential, as accurate weather prediction and effective domain adaptation are crucial for autonomous systems to adjust to dynamic environmental conditions.

Various tasks encountered in real-world surveillance can be addressed by determining posteriors (e.g. by Bayesian inference or machine learning), based on which critical decisions must be taken. However, the surveillance domain (acquisition device, operating conditions, etc.) is often unknown, which prevents any possibility of scene-specific optimization. In this paper, we define a probabilistic framework and present a formal proof of an algorithm for the unsupervised many-to-infinity domain adaptation of posteriors. Our proposed algorithm is applicable when the probability measure associated with the target domain is a convex combination of the probability measures of the source domains. It makes use of source models and a domain discriminator model trained off-line to compute posteriors adapted on the fly to the target domain. Finally, we show the effectiveness of our algorithm for the task of semantic segmentation in real-world surveillance. The code is publicly available at https://github.com/rvandeghen/MDA.

There exist many background subtraction algorithms to detect motion in videos. To help comparing them, datasets with ground-truth data such as CDNET or LASIESTA have been proposed. These datasets organize videos in categories that represent typical challenges for background subtraction. The evaluation procedure promoted by their authors consists in measuring performance indicators for each video separately and to average them hierarchically, within a category first, then between categories, a procedure which we name "summarization". While the summarization by averaging performance indicators is a valuable effort to standardize the evaluation procedure, it has no theoretical justification and it breaks the intrinsic relationships between summarized indicators. This leads to interpretation inconsistencies. In this paper, we present a theoretical approach to summarize the performances for multiple videos that preserves the relationships between performance indicators. In addition, we give formulas and an algorithm to calculate summarized performances. Finally, we showcase our observations on CDNET 2014.

Given a video sequence acquired from a fixed camera, the stationary background generation problem consists of generating a unique image estimating the stationary background of the sequence. During the IEEE Scene Background Modeling Contest (SBMC) organized in 2016, we presented the LaBGen-P method. In short, this method relies on a motion detection algorithm for selecting, for each pixel location, a given amount of pixel intensities that are most likely static by keeping the ones with the smallest quantities of motion. These quantities are estimated by aggregating the motion scores returned by the motion detection algorithm in the spatial neighborhood of the pixel. After this selection process, the background image is then generated by blending the selected intensities with a median filter. In our previous works, we showed that using a temporally-memoryless motion detection, detecting motion between two frames without relying on additional temporal information, leads our method to achieve the best performance. In this work, we go one step further by developing LaBGen-P-Semantic, a variant of LaBGen-P, the motion detection step of which is built on the current frame only by using semantic segmentation. For this purpose, two intra-frame motion detection algorithms, detecting motion from a unique frame, are presented and compared. Our experiments, carried out on the Scene Background Initialization (SBI) and SceneBackgroundModeling.NET (SBMnet) datasets, show that leveraging semantic segmentation improves the robustness against intermittent motions, background motions and very short video sequences, which are among the main challenges in the background generation field. Moreover, our results confirm that using an intra-frame motion detection is an appropriate choice for our method and paves the way for more techniques based on semantic segmentation.

Interactivity is one of the key challenges for immersive applications like gaming. Manufacturers have been working towards interfaces that are driven by a device (e.g. a Wiimote) or interfaces that are controlled by a camera with a subsequent computer vision module. Both approaches have unique advantages, but they do not permit to localize users in the scene with an appropriate accuracy. Therefore, we propose to use both a range camera and accurate range sensors to enable the interpretation of movements. This paper describes a platform that uses a range camera to acquire the silhouettes of users, regardless of illumination, and to improve the pose recovery with range information after some image processing steps. In addition, to circumvent the difficult process of calibration required to map range values to physical distances, we complete the system with several range laser sensors. These sensors are located in a horizontal plane, and measure distances up to a few centimeters. We combine all these measurements to obtain a localization map, used to locate users in the scene at a negligible computational cost. Our method fills a gap in 3D applications that require absolute positions.