The current sanitary emergency situation caused by COVID-19 has increased the interest in controlling flow of people indoor infrastructures, to ensure compliance with established security measures. Top view camera-based solutions have proven be an effective and non-invasive approach accomplish this task. Nevertheless, suffer from scalability problems: they cover limited range areas avoid dealing occlusions only work single camera scenarios. To overcome these problems, we present efficient...

Abstract Objective The study aimed to characterize morphological changes of the retinal microvascular network during progression diabetic retinopathy. Methods Publicly available images captured by a digital fundus camera from DIARETDB 1 and STARE databases were used. microvessels segmented using automatic method, vascular morphology was analyzed fractal parametrization such as box‐counting dimension, lacunarity, multifractals. Results results analysis affected ability segmentation method...

Curb detection is essential for environmental awareness in Automated Driving (AD), as it typically limits drivable and non-drivable areas. Annotated data are necessary developing validating an AD function. However, the number of public datasets with annotated point cloud curbs scarce. This paper presents a method detecting 3D sequence clouds captured from LiDAR sensor, which consists two main steps. First, our approach detects at each scan using segmentation deep neural network. Then,...

This paper introduces a web application for point cloud annotation that is used in the advanced driver assistance systems field. Apart from viewer, tool has an object viewer and timeline to define attributes of annotations video validate with corresponding images. The also describes several strategies we followed obtain correctly quickly: (i) memory management rendering large-scale clouds, (ii) coherent combination images annotations, (iii) content synchronization all parts (iv) automatic...

Face recognition provides a desirable solution for authentication and surveillance in Internet of Things platforms elderly care. However, its inclusion is challenging because the possibly reduced interaction capabilities users, high variety devices, need managing biometric data securely. Our approach relies on lightweight deep neural networks secure to guide users during interaction. An automated procedure selects appropriate inference engine, model configurations, batch size, based edge...

Nowadays, containerized freight transport is one of the most important transportation systems that undergoing an automation process due to Deep Learning success. However, it suffers from a lack annotated data in order incorporate state-of-the-art neural network models its systems. In this paper we present innovative methodology generate realistic, varied, balanced, and labelled dataset for visual inspection task containers dock environment. addition, validate with multiple tasks recurrently...

In this paper, we propose an approach to optimize the deployment of on-board video analytics for checking correct positioning luggage in aircraft cabins. The system consists embedded cameras installed on top cabin and a heterogeneous processor. Each camera covers multiple regions interest (i.e., seats or aisle sections) minimize number required. image region is processed by separate classification algorithm trained with expected kind visual appearance considering effect perspective lens...

This paper presents a new approach to 3D object detection that leverages the properties of data obtained by LiDAR sensor.State-of-the-art detectors use neural network architectures based on assumptions valid for camera images.However, point clouds from are fundamentally different.Most shared filter kernels extract features which do not take into account range dependent nature cloud features.To show this, different trained two splits KITTI dataset: close (points up 25 meters LiDAR) and...

Synthetic simulated environments are gaining popularity in the Deep Learning Era, as they can alleviate effort and cost of two critical tasks to build multi-camera systems for surveillance applications: setting up camera system cover use cases generating labeled dataset train required Neural Networks (DNNs).However, there no ready solve them all kind scenarios cases.Typically, 'ad hoc' built, which cannot be easily applied other contexts.In this work we present a methodology synthetic with...

In this paper, we tackle the problem of deploying face recognition (FR) solutions in heterogeneous Internet Things (IoT) platforms. The main challenges are optimal deployment deep neural networks (DNNs) high variety IoT devices (e.g., robots, tablets, smartphones, etc.), secure management biometric data while respecting users’ privacy, and design appropriate user interaction with facial verification mechanisms for all kinds users. We analyze different approaches to solving these propose a...

Gaze-annotated facial data is crucial for training deep neural networks (DNNs) gaze estimation. However, obtaining these labor-intensive and requires specialized equipment due to the challenge of accurately annotating direction a subject. In this work, we present generative framework create annotated by leveraging benefits labeled unlabeled sources. We propose Gaze-aware Compositional GAN that learns generate images from limited dataset. Then transfer model an domain take advantage diversity...

Building ADAS (Advanced Driver Assistance Systems) or AD (Autonomous Driving) vehicles implies the acquisition of large volumes data and a costly annotation process to create labeled metadata. Labels are then used for either ground truth composition (for test validation algorithms) set-up training datasets machine learning processes. In this paper we present 3D object tracking mechanism that operates on detections from point cloud sequences. It works in two steps: first an online phase which...