1,721,049 research outputs found
Outdoor People Detection in Low Resolution Thermal Images
No full textPresence detection is a main functionality to make our living spaces smarter and is implemented through several kinds of sensors and smart devices. Recent advancements in embedded systems market and technology enable the design of sophisticated solutions in a low-cost and scalable fashion. However, applications of presence detection, such as surveillance or occupancy detection, home automation or smart lighting are built for indoor scenarios. Therefore, many systems weaken their performance when applied outdoor, where ambient conditions have higher variability. In this work, we describe our exploratory study on people detection in outdoor scenarios by use of an 8×8 pixels resolution thermal sensor. We tested different techniques to extract the presence of a person crossing the detection area. We observed that signal to noise ratio depends on the difference between background and human body temperature. To address this, we collected a dataset spanning a wide range of background conditions and different user clothing and we used it to tune and evaluate the proposed detection techniques. As a possible solution, we propose to adapt the threshold with temperature, providing a regression curve to select it and demonstrate benefits against the use of a fixed threshold with all explored techniques
Convolutional Neural Network on Embedded Platform for People Presence Detection in Low Resolution Thermal Images
No full textDetection of human presence is a key feature in Human Computer Interaction. Solutions based on cameras are attractive, but require computer vision techniques to extract meaningful data, which can be expensive from a computational point of view. In this work, we present a new system that merges a low resolution thermal camera with advanced feature extraction techniques such as Convolutional Neural Networks. We demonstrate the possibility to adapt their execution to resource-constrained platform without significant loss of performance, by processing data on a 32-bit low power microcontroller, performing the classification on thermal video stream. It achieve 76.7% of accuracy in the microcontroller, requiring only 16.5 mW in continuous classification mode and using 6 kB of RAM
Quality-based registration refinement of airborne LiDAR and photogrammetric point clouds
No full textA big challenge in geodata processing is the seamless and accurate integration of airborne LiDAR (Light Detection And Ranging) and photogrammetric point clouds performed by properly considering their high variations in resolution and precision. In this paper we propose a new approach to co-register airborne point clouds acquired by LiDAR sensors and photogrammetric algorithms, assuming that only dense point clouds from both mapping methods are available, without LiDAR raw data nor flight trajectories. First, semantically segmented point clouds are quality-wise evaluated by assigning sensor-specific quality features to each 3D point. Then, these quality features are aggregated in order to assign a score to each 3D point based on its quality. Finally, using a voxel-based structure, a filtering step is performed to select only the best points used for the registration refinement. We assess the performance of the proposed method on two different case studies to demonstrate its advantages compared to a traditional ICP-based approach. The code of the implemented method is available at https://github.com/3DOM-FBK/HyRe
Geometric feature analysis for the classification of cultural heritage point clouds
Full text linkIn the last years, the application of artificial intelligence (Machine Learning and Deep Learning methods) for the classification of 3D point clouds has become an important task in modern 3D documentation and modelling applications. The identification of proper geometric and radiometric features becomes fundamental to classify 2D/3D data correctly. While many studies have been conducted in the geospatial field, the cultural heritage sector is still partly unexplored. In this paper we analyse the efficacy of the geometric covariance features as a support for the classification of Cultural Heritage point clouds. To analyse the impact of the different features calculated on spherical neighbourhoods at various radius sizes, we present results obtained on four different heritage case studies using different features configurations
People/Car Classification using an Ultra-Low-Power Smart Vision Sensor
No full textDeploying Internet of Things (IoT) in our cities will enable them to become smarter, thanks to the connection of everything everywhere, such as smart meters, street lighting, trash bin sensors, parking areas. However, a centralized-architecture approach, where all sensors and actuators send and receive data from the cloud, is not sustainable in terms of both the amount of data flooding from sensors to the cloud and the energy required to keep all these sensors alive. This is particularly true in the field of vision sensors, where the amount of data to be handled and transmitted can be high, while the real information we are interested in is possibly less "bulky" (e.g. a classification category or a feature). Data reduction is therefore desirable at the node level. This paper evaluates the use of a smart sensor, the FORENSOR sensor, which embeds motion detection in hardware, in a classification scenario. We achieve 87% accuracy, and we demonstrate the advantages of our sensor w.r.t frame-difference based ones. We discuss the classification algorithm chosen and we present the estimation of the power consumption, proving that the overall system consumes less than 2mW, thus being adequate for an IoT scenario
Di-segno, ricostruzione 3D e navigazione virtuale. Il racconto dell'utopia interrotta di Ferdinandopoli
No full textA San Leucio l'incompiuta città di Ferdinandopoli racconta due storie: le aspirazioni di un sovrano, Ferdinando IV, e la parziale realizzazione di un'utopia urbana. sfruttando le moderne metodologie
di rilievo fotogrammetrico digitale, lo studio ha permesso la realizzazione di un catalogo tridimensionale, metrico e colorimetrico, degli elementi ricorrenti nell' architettura costruita. Il confronto e l'integrazione tra i dati ottenuti e i risultati raggiunti dagli studi precedenti hanno fornito la base per la modellazione
30 del non costruito. l a ricostruzione virtuale è resa interattiva e navigabile, rendendo possibile la narrazione della storia incompiuta di un progetto illuminato
Neural network distillation on IoT platforms for sound event detection
No full textIn most classification tasks, wide and deep neural networks perform and generalize better than their smaller counterparts, in particular when they are exposed to large and heterogeneous training sets. However, in the emerging field of Internet of Things memory footprint and energy budget pose severe limits on the size and complexity of the neural models that can be implemented on embedded devices. The Student-Teacher approach is an attractive strategy to distill knowledge from a large network into smaller ones, that can fit on low-energy low-complexity embedded IoT platforms. In this paper, we consider the outdoor sound event detection task as a use case. Building upon the VGGish network, we investigate different distillation strategies to substantially reduce the classifier's size and computational cost with minimal performance losses. Experiments on the UrbanSound8K dataset show that extreme compression factors (up to 4.2 · 10−4 for parameters and 1.2 · 10−3 for operations with respect to VGGish) can be achieved, limiting the accuracy degradation from 75% to 70%. Finally, we compare different embedded platforms to analyze the trade-off between available resources and achievable accuracy
Analysis of Robust Implementation of an EMG Pattern Recognition Based Control
No full textControl of active hand prostheses is an open challenge. In fact, the advances in mechatronics made available prosthetic hands with multiple active degrees of freedom; however the predominant control strategies are still not natural for the user, enabling only few gestures, thus not exploiting the prosthesis potential. Pattern recognition and machine learning techniques can be of great help when applied to surface electromyography signals to offer a natural control based on the contraction of muscles corresponding to the real movements. The implementation of such approach for an active prosthetic system offers many challenges related to the reliability of data collected to train the classification algorithm. This paper focuses on these problems and propose an implementation suitable for an embedded system
Compact Recurrent Neural Networks for Acoustic Event Detection on Low-Energy Low-Complexity Platforms
Full text linkOutdoor acoustic event detection is an exciting research field but challenged by the need for complex algorithms and deep learning techniques, typically requiring many computational, memory, and energy resources. These challenges discourage IoT implementations, where an efficient use of resources is required. However, current embedded technologies and microcontrollers have increased their capabilities without penalizing energy efficiency. This paper addresses the application of sound event detection at the very edge, by optimizing deep learning techniques on resource-constrained embedded platforms for the IoT. The contribution is two-fold: firstly, a two-stage student-teacher approach is presented to make state-of-the-art neural networks for sound event detection fit on current microcontrollers; secondly, we test our approach on an ARM Cortex M4, particularly focusing on issues related to 8-bits quantization. Our embedded implementation can achieve 68% accuracy in recognition on Urbansound8k, not far from state-of-the-art performance, with an inference time of 125 ms for each second of the audio stream, and power consumption of 5.5 mW in just 34.3 kB of RAM
- …
