119 research outputs found
Improved Image Boundaries for Better Video Segmentation
No full textGraph-based video segmentation methods rely on superpixels as starting point. While most previous work has focused on the construction of the graph edges and weights as well as solving the graph partitioning problem, this paper focuses on better superpixels for video segmentation. We demonstrate by a comparative analysis that superpixels extracted from boundaries perform best, and show that boundary estimation can be significantly improved via image and time domain cues. With superpixels generated from our better boundaries we observe consistent improvement for two video segmentation methods in two different datasets
Perception pour véhicule urbain sans conducteur : conception et implémentation
No full textThe development of driverless vehicles capable of moving on urban roads could pro- vide important benefits in accidents reductions, life comfort and costs reductions. In this document we discuss how to create a perception system allowing robots to drive on roads, without the need to adapt the infrastructure, without requiring previous visits, and considering possible the presence of pedestrians and cars. We argue that the perception process is application specific and by nature needs to be able to deal with uncertainties in the knowledge of the world. We analyse the particular problem of perception for safe driving in the urban environments and propose a novel solution where the perception process is essentially seen as an optimization process. Also we propose that the perception process could benefit from collaboration between nearby vehicles. We examine this problem and provide a solution adapted to the con- straints encountered in the urban scenario. Here the core issue is formulated as a data association problem. Both the new perception and collaboration mechanism were integrated into a full-fledged driverless vehicle system. The results are supported by real world full scale experiments on our automated electric vehicles, the Cycabs.Le développement de véhicules sans conducteur capables de se déplacer sur des routes urbaines pourrait offrir des avantages importants dans la réduction des accidents, le confort et la réduction des coûts des déplacements. Dans ce document, nous discutons la manière de créer un système de perception permettant à un robot de conduire sur des routes sans devoir adapter l'infrastructure, sans avoir besoin de visites précédentes, et en prenant en compte la présence de piétons et d'autres voitures. Nous affirmons que le processus de perception est spécifique à l'application visée et que, par nature, il doit être capable de gérer les incertitudes dans la connaissance du monde. Nous analysons le problème de perception pour une conduite sûre dans les environnements urbains et proposons une solution où le processus de perception est formulé comme un processus d'optimisation. Nous pensons aussi que le processus de perception pourrait bénéficier de la collaboration entre véhicules à proximité. Nous examinons ce problème et proposons une solution adaptée aux contraintes rencontrés dans les zones urbaines. Ici la question centrale est formulée comme un problème d'association de données. La conception théorique du nouveau système de perception est validé par des expérimentations à pleine échelle sur notre véhicule électrique automatisé, le Cycab
Design of an urban driverless ground vehicle
No full textInternational audienceThis paper presents the design and implementation of a driverless car for populated urban environments. We propose a system that explicitly map the static obstacles, detects and track the moving obstacle, consider the unobserved areas, provide a motion plan with safety guarantees and executes it. All of it was implemented and integrated into a single computer maneuvering on real time an electric vehicle into an unvisited area with moving obstacles. The overview of the algorithms and some experimental results are presented
Synthetic Biology Platform for Sensing and Integrating Endogenous Transcriptional Inputs in Mammalian Cells
No full textSummaryOne of the goals of synthetic biology is to develop programmable artificial gene networks that can transduce multiple endogenous molecular cues to precisely control cell behavior. Realizing this vision requires interfacing natural molecular inputs with synthetic components that generate functional molecular outputs. Interfacing synthetic circuits with endogenous mammalian transcription factors has been particularly difficult. Here, we describe a systematic approach that enables integration and transduction of multiple mammalian transcription factor inputs by a synthetic network. The approach is facilitated by a proportional amplifier sensor based on synergistic positive autoregulation. The circuits efficiently transduce endogenous transcription factor levels into RNAi, transcriptional transactivation, and site-specific recombination. They also enable AND logic between pairs of arbitrary transcription factors. The results establish a framework for developing synthetic gene networks that interface with cellular processes through transcriptional regulators
Improving the care of adult sickle cell disease patients presenting with acute vaso-occlusive crisis to the emergency department via education
No full textPurpose: The purpose of this project was to improve pain management of sickle cell disease (SCD) patients presented to the emergency department (ED) with painful vaso-occlusive crisis (VOC).
Methodology: This project used a quasi-experimental pre-test post-test design. Registered nurses and other healthcare providers working in the adult ED participated in this project. A 30-min educational intervention on evidence-based pain management of SCD patients with VOC was carried out over a two-week period. A chart review was performed pre- and post- intervention to determine the impact of the intervention on pain management outcomes.
Results: Thirty RNs and four providers attended the educational sessions. The data analysis was based on review of 15 charts pre- and 12 charts post-intervention. There was no statistically significant improvement in assigning a higher triage level (p =0.482), starting a first dose of analgesics within 30 minutes after being triaged (p >0.05), reassessing pain within 30 minutes (p = 0.082), escalating a dose of pain medications (p = 0.765), and documenting pain at the end of the ED stay (0.542).
Implications for Practice: Future interventions should incorporate a pain management protocol in addition to educational sessions.DNPIncludes bibliographical reference
A Convnet for Non-maximum Suppression
No full textNon-maximum suppression (NMS) is used in virtually all state-of-the-art object detection pipelines. While essential object detection ingredients such as features, classifiers, and proposal methods have been extensively researched surprisingly little work has aimed to systematically address NMS. The de-facto standard for NMS is based on greedy clustering with a fixed distance threshold, which forces to trade-off recall versus precision. We propose a convnet designed to perform NMS of a given set of detections. We report experiments on a synthetic setup, and results on crowded pedestrian detection scenes. Our approach overcomes the intrinsic limitations of greedy NMS, obtaining better recall and precision
CityPersons: A Diverse Dataset for Pedestrian Detection
No full textConvnets have enabled significant progress in pedestrian detection recently, but there are still open questions regarding suitable architectures and training data. We revisit CNN design and point out key adaptations, enabling plain FasterRCNN to obtain state-of-the-art results on the Caltech dataset. To achieve further improvement from more and better data, we introduce CityPersons, a new set of person annotations on top of the Cityscapes dataset. The diversity of CityPersons allows us for the first time to train one single CNN model that generalizes well over multiple benchmarks. Moreover, with additional training with CityPersons, we obtain top results using FasterRCNN on Caltech, improving especially for more difficult cases (heavy occlusion and small scale) and providing higher localization quality
- …
