International audience ; The international tourism growth forces governments to make a big effort to improve the security of national borders. The compulsory passport stamping is used in guaranteeing the safekeeping of the entry point of the border. For each passenger, the border police must check the existence of exit stamps and/or the entry stamps of the country that the passenger visits, in all the pages of his passport. However, the systematic control considerably slows the operations of the border police. Protecting the borders from illegal immigrants and simplifying border checkpoints for law-abiding citizens and visitors is a delicate compromise. The purpose of this paper is to perform a flexible and scalable system that ensures faster, safer and more efficient stamp controlling. An automatic system of stamp extraction for travel documents is proposed. We incorporate several methods from the field of artificial intelligence, image processing and pattern recognition. At first, texture feature extraction is performed in order to find potential stamps. Next, image segmentation aimed at detecting objects of specific textures are employed. Then, isolated objects are extracted and classified using multi-layer perceptron artificial network. Promising results are obtained in terms of accuracy, with a maximum average of 0.945 among all the images, improving the performance of MLP neural network in all cases.
International audience ; The international tourism growth forces governments to make a big effort to improve the security of national borders. The compulsory passport stamping is used in guaranteeing the safekeeping of the entry point of the border. For each passenger, the border police must check the existence of exit stamps and/or the entry stamps of the country that the passenger visits, in all the pages of his passport. However, the systematic control considerably slows the operations of the border police. Protecting the borders from illegal immigrants and simplifying border checkpoints for law-abiding citizens and visitors is a delicate compromise. The purpose of this paper is to perform a flexible and scalable system that ensures faster, safer and more efficient stamp controlling. An automatic system of stamp extraction for travel documents is proposed. We incorporate several methods from the field of artificial intelligence, image processing and pattern recognition. At first, texture feature extraction is performed in order to find potential stamps. Next, image segmentation aimed at detecting objects of specific textures are employed. Then, isolated objects are extracted and classified using multi-layer perceptron artificial network. Promising results are obtained in terms of accuracy, with a maximum average of 0.945 among all the images, improving the performance of MLP neural network in all cases.
International audience ; The international tourism growth forces governments to make a big effort to improve the security of national borders. The compulsory passport stamping is used in guaranteeing the safekeeping of the entry point of the border. For each passenger, the border police must check the existence of exit stamps and/or the entry stamps of the country that the passenger visits, in all the pages of his passport. However, the systematic control considerably slows the operations of the border police. Protecting the borders from illegal immigrants and simplifying border checkpoints for law-abiding citizens and visitors is a delicate compromise. The purpose of this paper is to perform a flexible and scalable system that ensures faster, safer and more efficient stamp controlling. An automatic system of stamp extraction for travel documents is proposed. We incorporate several methods from the field of artificial intelligence, image processing and pattern recognition. At first, texture feature extraction is performed in order to find potential stamps. Next, image segmentation aimed at detecting objects of specific textures are employed. Then, isolated objects are extracted and classified using multi-layer perceptron artificial network. Promising results are obtained in terms of accuracy, with a maximum average of 0.945 among all the images, improving the performance of MLP neural network in all cases.
International audience ; Influenced by the field of Computer Vision, Generative Adversarial Networks (GANs) are often adopted for the audio domain using fixed-size two-dimensional spectrogram representations as the "image data". However, in the (musical) audio domain, it is often desired to generate output of variable duration. This paper presents VQCPC-GAN, an adversarial framework for synthesizing variablelength audio by exploiting Vector-Quantized Contrastive Predictive Coding (VQCPC). A sequence of VQCPC tokens extracted from real audio data serves as conditional input to a GAN architecture, providing step-wise time-dependent features of the generated content. The input noise z (characteristic in adversarial architectures) remains fixed over time, ensuring temporal consistency of global features. We evaluate the proposed model by comparing a diverse set of metrics against various strong baselines. Results show that, even though the baselines score best, VQCPC-GAN achieves comparable performance even when generating variable-length audio. Numerous sound examples are provided in the accompanying website, 1 and we release the code for reproducibility. 2 Index Terms-Generative Adversarial Networks, Audio Synthesis, Vector-Quantized Contrastive Predictive Coding * Nistal received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 765068. 1 sonycslparis.github.io/vqcpc-gan.io 2 github.com/SonyCSLParis/vqcpc-gan
International audience ; Influenced by the field of Computer Vision, Generative Adversarial Networks (GANs) are often adopted for the audio domain using fixed-size two-dimensional spectrogram representations as the "image data". However, in the (musical) audio domain, it is often desired to generate output of variable duration. This paper presents VQCPC-GAN, an adversarial framework for synthesizing variablelength audio by exploiting Vector-Quantized Contrastive Predictive Coding (VQCPC). A sequence of VQCPC tokens extracted from real audio data serves as conditional input to a GAN architecture, providing step-wise time-dependent features of the generated content. The input noise z (characteristic in adversarial architectures) remains fixed over time, ensuring temporal consistency of global features. We evaluate the proposed model by comparing a diverse set of metrics against various strong baselines. Results show that, even though the baselines score best, VQCPC-GAN achieves comparable performance even when generating variable-length audio. Numerous sound examples are provided in the accompanying website, 1 and we release the code for reproducibility. 2 Index Terms-Generative Adversarial Networks, Audio Synthesis, Vector-Quantized Contrastive Predictive Coding * Nistal received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 765068. 1 sonycslparis.github.io/vqcpc-gan.io 2 github.com/SonyCSLParis/vqcpc-gan
International audience ; Safran has been working for several years on autonomy of vehicles. Whether it is airborne with the UAV Patroller, or on the ground with the military vehicle eRider and with the civilian autonomous car in cooperation with Valeo. This paper focuses on the use of visual information to improve the localisation of the car. More precisely, it presents, from a theoretical point of view, the different kind of visual information that can be used in the navigation filter to improve the localisation of the car and the corresponding hybridisation. The efficiency of these hybridisation are evaluated one by one on simulated and/or on real data.
International audience ; Safran has been working for several years on autonomy of vehicles. Whether it is airborne with the UAV Patroller, or on the ground with the military vehicle eRider and with the civilian autonomous car in cooperation with Valeo. This paper focuses on the use of visual information to improve the localisation of the car. More precisely, it presents, from a theoretical point of view, the different kind of visual information that can be used in the navigation filter to improve the localisation of the car and the corresponding hybridisation. The efficiency of these hybridisation are evaluated one by one on simulated and/or on real data.
International audience ; Development and deployment of underwater acoustic (UWA) sensor networks have emerged as an efficient solution for many applications related to marine environment e.g., scientific, industrial, and military operations. The main tasks to execute are remote sensing and monitoring. Grid topology with multihop relaying is very useful, since it allows a wide area coverage as well as long distance data transmission. In this paper, we investigate architectures where sensing and monitoring data are forwarded over multiple lines independently. We are interested in transmission schedules which maximize network throughput. We prove that an optimal schedule is necessarily per-node fair. We also derive the upper bound. Furthermore, we present a low-complexity algorithm to find schedules achieving the upper bound, regardless of the size of the network.
International audience ; Development and deployment of underwater acoustic (UWA) sensor networks have emerged as an efficient solution for many applications related to marine environment e.g., scientific, industrial, and military operations. The main tasks to execute are remote sensing and monitoring. Grid topology with multihop relaying is very useful, since it allows a wide area coverage as well as long distance data transmission. In this paper, we investigate architectures where sensing and monitoring data are forwarded over multiple lines independently. We are interested in transmission schedules which maximize network throughput. We prove that an optimal schedule is necessarily per-node fair. We also derive the upper bound. Furthermore, we present a low-complexity algorithm to find schedules achieving the upper bound, regardless of the size of the network.
International audience ; Development and deployment of underwater acoustic (UWA) sensor networks have emerged as an efficient solution for many applications related to marine environment e.g., scientific, industrial, and military operations. The main tasks to execute are remote sensing and monitoring. Grid topology with multihop relaying is very useful, since it allows a wide area coverage as well as long distance data transmission. In this paper, we investigate architectures where sensing and monitoring data are forwarded over multiple lines independently. We are interested in transmission schedules which maximize network throughput. We prove that an optimal schedule is necessarily per-node fair. We also derive the upper bound. Furthermore, we present a low-complexity algorithm to find schedules achieving the upper bound, regardless of the size of the network.
Satellite navigation is a promising technology for terrestrial applications that requires the monitoring of the vehicle position thanks to costly ground infrastructures. In the rail domain, the European train control system (ETCS) relies on a combination of radio beacons (that provides information of absolute position) and odometry to propagate the position between two balises groups. The use of Global Navigation Satellite Systems (GNSS) in ETCS has been proposed in order to reduce the amount of beacons. In the road domain, the GNSS is one of the technologies recommended by the European Union directive for Electronic Toll Collection (ETC), and GNSS based ETC systems already exists for heavy good transportation in Germany (Toll Collect) and Slovakia (MYTO). For these two applications that are either safety critical (train control) or liability critical (toll collection), it is not acceptable to estimate the position of the vehicle with a large error without warning the system within a sufficiently short delay. It is firstly necessary to define the operational requirements for the navigation system for these terrestrial applications. This kind of problematic has already been handled in the context of civil aviation which is also a critical application, but currently, the operational requirements associated to GNSS are not standardized for the train control and ETC. Based on the model of civil aviation, a state of the art of possible requirements for train control and electronic toll collection is proposed. For train control, a solution based on redundant independent GNSS constellations has been proposed in order to relax the integrity risk requirements on each sub constellation. In the case of ETC, the requirements will depend on the case of study and are indirectly imposed by the toll charger. For terrestrial applications, the vehicles are likely to operate in constrained environment (including urban environment). In urban environment, the performances of GNSS are highly degraded due to multipath interference, tracking of non-line-of-sights and masking effects. These phenomena are likely to degrades the accuracy, integrity, availability and continuity of the GNSS based positioning system. It is proposed to augment the solution proposed for each application by integrating measurements from a six axis inertial measurement unit which are insensitive to the receiver surrounding environment. Integrating information from other sensors such as a track database for train control or odometry for toll collection is investigated. The nominal error models and fault modes of the sensors are then studied. The nominal error models will be used to weight the measurements in the fusion algorithm and to test the performances of the fusion algorithm by realistic simulations. In particular, the characterization of the distribution and the modelling of the errors dues to multipath and non-line-of-sights in urban environment is studied on simulator and on a data collection campaign conducted in Toulouse downtown (France) and its surroundings. The extended Kalman filter used to fuse the GNSS measurements and the measurements from other sensors are then presented. A tight coupling architecture in closed loop is presented as it is the most adapted to the cases of study. The integration of a track database in the solution is discussed in the case of train control. The extension of the solution to the multi-constellation case is also presented. The solutions have been validated and tested on a simulator as well as in real condition in Toulouse downtown. It is shown that additional sensors such as track database or wheel speed sensors enable to limit the drift of the position error in costing/degraded constellation condition. Then, it is proposed to improve the robustness and the reliability of the GNSS measurements in urban area by developing multipath detection algorithms at the signal processing level. A detection algorithm based on the real time analysis of the correlation function is proposed. This algorithm aims at assisting the integrity monitoring algorithm upstream by protecting it against the faults due to multipath with large amplitudes. However, this algorithm does not protect against non-line-of-sights that can lead to integrity failures as this phenomenon is not associated with any abnormal distortion. Several methods based on the elevations of the satellites, the signal to noise ratio, or the coherence of the measurements based on the comparison with non GNSS sensors measurements have been studied in order to protect the solution against this phenomenon. Two snapshots integrity monitoring algorithms adapted to the Kalman filter are presented. Finally, the performances of these algorithms are tested on the data collected in Toulouse downtown and surroundings. The improvement obtained by assisting the integrity monitoring algorithm upstream by testing the quality of the measurement is quantized (with respect to a simple inflation of the nominal models in urban environments). ; La navigation par satelolite est une technologie prometteuse pour les applications terrestresqui necessitent le controle d'un véhicule à partir d'infrastructures sol couteuses.
Satellite navigation is a promising technology for terrestrial applications that requires the monitoring of the vehicle position thanks to costly ground infrastructures. In the rail domain, the European train control system (ETCS) relies on a combination of radio beacons (that provides information of absolute position) and odometry to propagate the position between two balises groups. The use of Global Navigation Satellite Systems (GNSS) in ETCS has been proposed in order to reduce the amount of beacons. In the road domain, the GNSS is one of the technologies recommended by the European Union directive for Electronic Toll Collection (ETC), and GNSS based ETC systems already exists for heavy good transportation in Germany (Toll Collect) and Slovakia (MYTO). For these two applications that are either safety critical (train control) or liability critical (toll collection), it is not acceptable to estimate the position of the vehicle with a large error without warning the system within a sufficiently short delay. It is firstly necessary to define the operational requirements for the navigation system for these terrestrial applications. This kind of problematic has already been handled in the context of civil aviation which is also a critical application, but currently, the operational requirements associated to GNSS are not standardized for the train control and ETC. Based on the model of civil aviation, a state of the art of possible requirements for train control and electronic toll collection is proposed. For train control, a solution based on redundant independent GNSS constellations has been proposed in order to relax the integrity risk requirements on each sub constellation. In the case of ETC, the requirements will depend on the case of study and are indirectly imposed by the toll charger. For terrestrial applications, the vehicles are likely to operate in constrained environment (including urban environment). In urban environment, the performances of GNSS are highly degraded due to multipath interference, ...
International audience ; The purpose of this presentation is to describe why and how psychoacoustic models are used to design speech enhancement systems capable of improving speech intelligibility in presence of noise (for mobile communication applications, for instance) or capable of denoising sufficiently well noisy speech signals so as to improve the recognition rate of some automatic speech recognizer (for instance, robot monitoring in noisy environment, hand-free applications on board of vehicles, military fastjets and helicopters). To begin with, standard methods aimed at denoising speech signals are performed in the spectral domain without taking into account the perceptual characteristics of the speech signal to enhance. They succeed in improving the Signal to Noise Ratio (SNR) but return annoying and unpleasant residual noise known as musical noise. In the last few decades, psychoacoustic models have then attracted a great deal of interest. The objective is to improve the perceptual quality of the enhanced speech signal. The psychoacoustic model is used to control the enhancement process in order to find the best trade-off between noise reduction, residual noise and speech distortion. The masking phenomenon is the main human auditory system property which is used to design perceptually motivated speech enhancement systems.
International audience ; The purpose of this presentation is to describe why and how psychoacoustic models are used to design speech enhancement systems capable of improving speech intelligibility in presence of noise (for mobile communication applications, for instance) or capable of denoising sufficiently well noisy speech signals so as to improve the recognition rate of some automatic speech recognizer (for instance, robot monitoring in noisy environment, hand-free applications on board of vehicles, military fastjets and helicopters). To begin with, standard methods aimed at denoising speech signals are performed in the spectral domain without taking into account the perceptual characteristics of the speech signal to enhance. They succeed in improving the Signal to Noise Ratio (SNR) but return annoying and unpleasant residual noise known as musical noise. In the last few decades, psychoacoustic models have then attracted a great deal of interest. The objective is to improve the perceptual quality of the enhanced speech signal. The psychoacoustic model is used to control the enhancement process in order to find the best trade-off between noise reduction, residual noise and speech distortion. The masking phenomenon is the main human auditory system property which is used to design perceptually motivated speech enhancement systems.
International audience ; Under the 2004 Agreement on the Promotion, Provision, and Use of GALILEO and GPS Satellite-Based Navigation Systems and Related Applications, the member states of the European Union and the United States agreed on working together, intensifying thus the cooperation on interoperability and compatibility issues between GALILEO and GPS. Among other topics, one important focus was the E1/L1 frequency band, centred at 1575.42 MHz, where the GALILEO Open Service (OS) signal and the modernized L1 civil (L1C) signal are going to be transmitted along with many other RNSS signals. The opportunity to design new signals in this preeminent radionavigation frequency band has significant importance to future users worldwide. Recent joint efforts by United States and European experts have identified MBOC (multiplexed BOC) [1][2][3][4][5] as a promising joint solution for E1 OS and L1C, along with multiple sets of spreading waveforms that yield this optimized spectrum. The resulting optimized E1 OS and L1C spreading modulations enable receivers to obtain significantly better performance in multipath than with previously considered spreading modulations, along with other potential benefits. The optimized spreading modulation provides considerable flexibility for receiver designers, and simpler receivers that employ only BOC(1,1)-based processing, experience very modest performance degradation, compared to the baseline BOC(1,1) spreading modulation.