This paper deals simultaneously with the trajectory estimation and map reconstruction by means of a stereo-calibrated vision system evolving in a large-scale unknown environment. This problem is widely known as Visual SLAM. Our proposal optimizes the execution time of the VSLAM framework while preserving its localization accuracy. The contributions of this paper are structured as follows. First, a novel VSLAM approach based on a ``Weighted Mean'' of multiple neighbor poses is detailed and is denoted as HOOFR SLAM. This approach provides a localization estimate after computing the camera poses (6-DOF rigid transformation) from the current image frame to previous neighbor frames. Taking advantage of the camera motion, we conjointly incorporate two types of stereo modes: ``Static Stereo'' mode (SS) through the fixed-baseline of left-right cameras setup along with the ``Temporal Multi-view Stereo'' mode (TMS). Moreover, instead of computing beforehand the disparity of SS mode for all key-points set, the disparity map in scale estimation step is limited to the inliers of the TMS mode so as to reduce the computational cost. This strategy is suitable to be parallelized on a multiprocessor architecture and exhibits a competitive performance with the other state-of-the-art strategies in many real datasets. Second, we report a hardware-software mapping of the proposed VSLAM approach. To this end, a heterogeneous CPU-GPU architecture-based vision system is considered. Third, a thorough and extensive experimental evaluation of our algorithm implemented on an automotive architecture (the NVIDIA Tegra TX1 system) is studied and analyzed. We report hence the localization and timing results through experiments on five well-known public stereo SLAM datasets: KITTI, Malaga, Oxford, MRT, and St_Lucia datasets.

Motorcycle drivers are considered among the most vulnerable road users, as attested by the number of crashes increasing every year. The significant part of the fatalities relates to “single vehicle” loss of control in bends. During this investigation, a system based on an instrumented multi-sensor platform and an algorithmic study was developed to accurately reconstruct motorcycle trajectories achieved when negotiating bends. This system is used by the French Gendarmerie in order to objectively evaluate and to examine the way riders take their bends in order to better train riders to adopt a safe trajectory and to improve road safety. Data required for the reconstruction are acquired using a motorcycle that has been fully instrumented (in VIROLO++ Project) with several redundant sensors (reference sensors and low-cost sensors) which measure the rider actions (roll, steering) and the motorcycle behavior (position, velocity, acceleration, odometry, heading, and attitude). The proposed solution allowed the reconstruction of motorcycle trajectories in bends with a high accuracy (equal to that of fixed point positioning). The developed algorithm will be used by the French Gendarmerie in order to objectively evaluate and examine the way riders negotiate bends. It will also be used for initial training and retraining in order to better train riders to learn and estimate a safe trajectory and to increase the safety, efficiency and comfort of motorcycle riders.

In this data article, we will present the data coming from 3D Inertial Measurement Unit (3-accelerometers and 3-gyroscopes sensors) mounted on the motorcycle collected during a motorcycle's falls experiments. Developing a motorcycle's fall events detection algorithms is a very challenging task because the motorcycle falling is multi-factorial and is strongly influenced by many unknown factors. To solve this issue, one solution can be to use a data-set collected during controlled experiments, knowing that the real motorcycle falls cannot be replicated, a stuntman can be chosen to be as close to reality as possible. The experiments have been conducted based on predefined scenarios such as: fall in a curve, fall on a slippery straight road section, fall with leaning of the motorcycle ''intentional manoeuvre'' and fall in a roundabout. These scenarios have been designed based on realistic falls. Other experiments have been conducted under different extreme driving situations. These extreme manoeuvres were carried out on track by professional riders. The purpose of performing these manoeuvres was to obtain a dataset describing the limit handling behaviour.

Fixed-wing Unmanned Aerial Vehicles (UAVs) are a special class of UAVs which present many advantages notably long range of action. Whereas, design of this kind of UAVs requires heavy logistics like outdoor tests, runways, and experimented pilots. These constraints reverberate on the design of embedded systems for fixed-wing UAVs. Because static tests are not representative, this paper proposes a practical approach to evaluate an embedded system on an appropriate vehicle emulating the dynamic model of a fixed-wing aircraft. For that, a comparison between the dynamic model of fixed-wing aircraft, tank-type mobile robot, and a bicycle is achieved. We show that, contrary to trend in literature, a mobile robot is not the optimal choice to emulate a fixed-wing UAV. Indeed, supposing a motion without slip (and a constant altitude for the aircraft), translation models of the three vehicles are under the form of Dubin car model. Whereas, translation and rotation velocities of tank-type mobile robot are coupled (while it is not the case for the aircraft where propulsion and turning are actuated separately). This constraint defines an allowed kinematic zone which limits the emulation of a fixed wing airplane. In the other hand, in bicycle model “bank to turn effect” is similar to the one observed in fixed-wing aircraft model. Furthermore, both models are not defined when the translation velocity tends to zero (stalling effect). As a conclusion, we propose to use mobile robot to test the navigation layer, and the bicycle to evaluate the sensor processing layer of an embedded system based fixed-wing UAVs applications.