Valentin Magnier

his paper deals with real-time obstacle detection and track- ing using multi-layer LIDAR data. We present two algorithms to cluster raw data coming from LIDAR sensors. The rst algorithm is based on a dynamic clustering approach while the second one relies on the con- nectivity between the laser impacts. Both algorithms take into account the inaccuracy and the uncertainty of the data sources. We propose a tracking approach based on the belief theory to estimate the dynamic state of the detected… Voir plus

his paper deals with real-time obstacle detection and track- ing using multi-layer LIDAR data. We present two algorithms to cluster raw data coming from LIDAR sensors. The rst algorithm is based on a dynamic clustering approach while the second one relies on the con- nectivity between the laser impacts. Both algorithms take into account the inaccuracy and the uncertainty of the data sources. We propose a tracking approach based on the belief theory to estimate the dynamic state of the detected objects in order to predict their future maneuvers. The objects are then ltered using an intelligent ROI that depends on a dynamic evolution area computed from proprioceptive information of the ego-vehicle. We evaluate and validate the whole chained process on real data-sets. Voir moins

In Autonomous driving applications, the LIDAR is becoming one of the key sensors for the perception of the environment. Indeed its work principle which is based on distance ranging using a laser beam scanning the environment allows highly accurate measurements. Among sensors commonly used in autonomous driving applications, which are cameras, RADARs and LIDARs, the LIDAR is the most suited to estimate the shape of objects. However, for the moment, LIDARs dedicated to pure automotive application… Voir plus

In Autonomous driving applications, the LIDAR is becoming one of the key sensors for the perception of the environment. Indeed its work principle which is based on distance ranging using a laser beam scanning the environment allows highly accurate measurements. Among sensors commonly used in autonomous driving applications, which are cameras, RADARs and LIDARs, the LIDAR is the most suited to estimate the shape of objects. However, for the moment, LIDARs dedicated to pure automotive application have only up to four measurement layers (4 laser beams scanning the environment at different height). Hence objects detection algorithm have to rely on very few layers to detected and classify the type of objects perceived on the road scene, that makes them specific. In this paper we will present an Detection and Tracking of Moving Objects (DATMO) algorithm featuring an object-type classification based on the belief theory. This algorithm is specific to automotive application therefore, the classification of perceived vehicles is between bike, car and truck. At the end of this paper we will present an application of this algorithm in real-world context. Voir moins

For several years, researches dedicated to autonomous driving are growing and a great number of applications and algorithms have been developed for embedded ADAS. In a part of these very critical applications, the tracking of the road's obstacles is one of the key elements. The tracking is built around a stage of measurement to track association. It consists in assessing the similarity between two sets of multi-dimensional state vectors characterizing the track and the measurement. For this… Voir plus

For several years, researches dedicated to autonomous driving are growing and a great number of applications and algorithms have been developed for embedded ADAS. In a part of these very critical applications, the tracking of the road's obstacles is one of the key elements. The tracking is built around a stage of measurement to track association. It consists in assessing the similarity between two sets of multi-dimensional state vectors characterizing the track and the measurement. For this application, but not only, we propose in this work a multi-criteria similarity operator based on Belief Theory allowing to take into account uncertainties and imperfections on the data. Partial data availability, conflict and ambiguities between state vectors are managed. The output of this operator provides not only the similarity but also the dissimilarity level, the amount of conflict and ambiguity. This operator is applied to relevant use-cases to highlight its benefits over the standard similarity operators. Voir moins

This paper presents a Multi-Hypotheses Tracking (MHT) approach that allows solving ambiguities that arise with previous methods of associating targets and tracks within a highly volatile vehicular environment. The previous approach based on the Dempster–Shafer Theory assumes that associations between tracks and targets are unique; this was shown to allow the formation of ghost tracks when there was too much ambiguity or conflict for the system to take a meaningful decision. The MHT algorithm… Voir plus

This paper presents a Multi-Hypotheses Tracking (MHT) approach that allows solving ambiguities that arise with previous methods of associating targets and tracks within a highly volatile vehicular environment. The previous approach based on the Dempster–Shafer Theory assumes that associations between tracks and targets are unique; this was shown to allow the formation of ghost tracks when there was too much ambiguity or conflict for the system to take a meaningful decision. The MHT algorithm described in this paper removes this uniqueness condition, allowing the system to include ambiguity and even to prevent making any decision if available data are poor. We provide a general introduction to the Dempster–Shafer Theory and present the previously used approach. Then, we explain our MHT mechanism and provide evidence of its increased performance in reducing the amount of ghost tracks and false positive processed by the tracking system. Voir moins

This paper aims to present how the Belief Theory (also known as the Dempster-Shafer theory) can be relevant to implement powerful tracking systems. As the Belief theory belongs to the group of information-theories, it is very suitable for solving the track-to-target association problem which is one of the main issue of tracking systems. The data association problem is about associating the measurement at a given time-stamp with the objects that are being tracked along the time. In this paper… Voir plus

This paper aims to present how the Belief Theory (also known as the Dempster-Shafer theory) can be relevant to implement powerful tracking systems. As the Belief theory belongs to the group of information-theories, it is very suitable for solving the track-to-target association problem which is one of the main issue of tracking systems. The data association problem is about associating the measurement at a given time-stamp with the objects that are being tracked along the time. In this paper, we present two methods based on the belief theory that can be used to improve the data association reliability. After that we propose as an example of tracking implementation and discuss the effect of using a multi-criteria track-to-target association algorithm.
Published in: Voir moins

In this paper a new algorithm for multi-targets tracking for roadway environment is proposed. This new approach is based on two parallel tracking stages. Its objective is to improve associations between targets and tracks by avoiding wrong associations which can cause errors on track's path determination. Another interesting point of the proposed approach lies in the fact that the two trackings stages are operated together only when association ambiguities are detected. otherwise, only one… Voir plus

In this paper a new algorithm for multi-targets tracking for roadway environment is proposed. This new approach is based on two parallel tracking stages. Its objective is to improve associations between targets and tracks by avoiding wrong associations which can cause errors on track's path determination. Another interesting point of the proposed approach lies in the fact that the two trackings stages are operated together only when association ambiguities are detected. otherwise, only one tracking is used. This mechanism leads to save computational resources. This contribution comes after previous works achieved at the LIVIC (Laboratory on interactions between vehicles, road network and drivers) regarding to Multi-Hypothesis Tracking (MHT) using the Dempster-Shafer Theory. These previous works discussed the potential interest of considering at the same time multi-hypothesis solutions instead of mono-hypothesis ones. This new approach is more focused on the identification of ambiguities, and runs simultaneously two tracking stages in order to solve these ambiguities thanks to the Dempster-Shafer multi-criteria association rules. The paper will therefore explain quickly the basis of the MHT and then describe the Dual Tracking Ambiguities' Solving (DTAS) algorithm. Finally, a relevant case of study showing the interest of the DTAS will be discussed. Voir moins