Xavier Savatier

Robust Radial Face Detection for Omnidirectional Vision

Bio-inspired and non-conventional vision systems are highly researched topics. Among them, omnidirectional vision systems have demonstrated their ability to significantly improve the geometrical interpretation of scenes. However, few researchers have investigated how to perform object detection with such systems. The existing approaches require a geometrical transformation prior to the interpretation of the picture. In this paper, we investigate what must be taken into account and how to process omnidirectional images provided by the sensor. We focus our research on face detection and highlight the fact that particular attention should be paid to the descriptors in order to successfully perform face detection on omnidirectional images. We demonstrate that this choice is critical to obtaining high detection rates. Our results imply that the adaptation of existing object-detection frameworks, designed for perspective images, should be focused on the choice of appropriate image descriptors in the design of the object-detection pipeline.

A Study of Vicon System Positioning Performance

Motion capture setups are used in numerous fields. Studies based on motion capture data can be found in biomechanical, sport or animal science. Clinical science studies include gait analysis as well as balance, posture and motor control. Robotic applications encompass object tracking. Today’s life applications includes entertainment or augmented reality. Still, few studies investigate the positioning performance of motion capture setups. In this paper, we study the positioning performance of one player in the optoelectronic motion capture based on markers: Vicon system. Our protocol includes evaluations of static and dynamic performances. Mean error as well as positioning variabilities are studied with calibrated ground truth setups that are not based on other motion capture modalities. We introduce a new setup that enables directly estimating the absolute positioning accuracy for dynamic experiments contrary to state-of-the art works that rely on inter-marker distances. The system performs well on static experiments with a mean absolute error of 0.15 mm and a variability lower than 0.025 mm. Our dynamic experiments were carried out at speeds found in real applications. Our work suggests that the system error is less than 2 mm. We also found that marker size and Vicon sampling rate must be carefully chosen with respect to the speed encountered in the application in order to reach optimal positioning performance that can go to 0.3 mm for our dynamic study.

Fusion of Omnidirectional and PTZ Cameras for Face Detection and Tracking

Many applications for mobile robot authentication require to be able to explore a large field of view with high resolution. The proposed vision system is composed of a catadioptric sensor for full range monitoring and a pan tilt zoom (PTZ) camera leading to an innovative sensor, able to detect and track any moving objects at a higher zoom level. In our application, the catadioptric sensor is calibrated and used to detect and track regions of interest (ROIs) within its 360 degree field of view (FOV), especially face regions. Using a joint calibration strategy, the PTZ camera parameters are automatically adjusted by the system in order to detect and track the face ROI within a higher resolution.

An omnidirectional stereoscopic system for mobile robot navigation

This paper proposes a scheme for a 3D metric reconstruction of the environment of a mobile robot. We first introduce the advantages of a catadioptric stereovision sensor for autonomous navigation and how we have designed it with respect to the Single Viewpoint constraint. For applications such as path generation, the robot needs a metric reconstruction of its environment, so calibration of the sensor is required. After justification of the chosen model, a calibration method to obtain the model parameters and the relative pose of the two catadioptric sensors is presented. Knowledge of all the sensor parameters yields the 3D metric reconstruction of the environment by triangulation. Tools for calibration and relative pose estimation are presented and are available on the authorpsilas Web page. The entire process has been evaluated using real data.

A direct approach for face detection on omnidirectional images

Catadioptric sensors offer abilities unexploited so far. This is especially true for face detection, and more generally, object detection. This paper presents our results of a direct approach to tackle face detection on catadioptric images. Despite no geometrical transformations, we are able to successfully apply our detector on distorted images. We expose a new method to synthesize large omnidirectional images database. Inspired from regular face detection training schemes, our method makes use of newly introduced polygonal Haar-like features. First tests demonstrated that our approach gives good performance and at the same time speeds up the detection process.

Fast and robust vehicle positioning on graph-based representation of drivable maps

In this paper, we propose a car positioning approach that does not rely on GPS. We propose to use car wheel speeds and road maps in order to achieve robust positioning of the vehicle. The vehicle positioning is achieved by applying particle filtering on a graph-based representation of a road map. We show that the vehicle positioning is feasible and robust with these two inputs at a really low computational cost. We achieve car positioning with an averaged 5 m accuracy within a 100 km drivable road map on a 12 km sequence.

Real-time 3D reconstruction for mobile robot using catadioptric cameras

This paper presents a 3-second 3D reconstruction algorithm able to process a dense geometric approximation of the surrounding environment. Image acquisition is done by a stereoscopic panoramic system with two color catadioptric cameras mounted on a mobile robot. An algorithm running on a Graphical Processing Unit (GPU) processes the 3D reconstruction in real-time. As the camera system moves, new views of the scene are used to improve the model of the scene thanks to an incremental algorithm. Then, the performance of our approach is evaluated using a synthetic image sequence.

Study of susceptibility of an MCU control system in the automotive field

This paper introduces a technique to detect the EMI environment of MCU relying on itself. The measurement setup is explained in detail. Furthermore, based on measurement, an output signal data library and susceptibility results are achieved. Finally, we present a technique which can detect disturbance and PMS (program multi-switching) system which can increase the reliability of MCU control DC Motor system and guarantee system speed. The final objective of the work is to develop software technique to increase the reliability of MCU system in the automotive field.

3D real-time human action recognition using a spline interpolation approach

This paper presents a novel descriptor based on skeleton information provided by RGB-D videos for human action recognition. These features are obtained, considering the motion as continuous trajectories of skeleton joints. With the discrete information of skeleton joints position, a cubic-spline interpolation is applied to joints position, velocity and acceleration components. The training and classification steps are done using a linear SVM. In the literature, many human motion descriptors based on RGB-D cameras had already been proposed with good accuracy, but with a high computational time. The main interest of this proposed approach is its ability to calculate human motion descriptors with a low computation cost while such a descriptor leads to an acceptable accuracy of recognition. Thus, this approach can be adapted to human computer interaction applications. For the purpose of validation, we apply our method to the challenging benchmark MSR-Action3D and introduce a new indicator which is the ratio between accuracy and execution time per descriptor. Using this criterion, we show that our algorithm outperforms the state-of-art methods in terms of the combined information of rapidity and accuracy.

Wheel Odometry-Based Car Localization and Tracking on Vectorial Map

In this paper, we present a car self-localization approach based on free inputs. We propose to use wheel speeds, which is available on most car through the CAN bus, and community developed road maps. A particle filter framework is used to achieve self-localization on a graph-based representation of a road map. Our results suggests that self-localization and tracking are feasible with these two inputs at a really low computational cost. Car self-localization is achieved with an averaged 5 m accuracy within a 100 km drivable road map on a 12 km sequence.