Skin cancer is a major health problem. There are several techniques to help diagnose skin lesions from a captured image. Computer-aided diagnosis (CAD) systems operate on single images of skin lesions, extracting lesion features to further classify them and help the specialists. Accurate feature extraction, which further depends on precise lesion segmentation, is key for the performance of these systems. In this paper, we present a skin lesion segmentation algorithm based on a novel adaptation of superpixels techniques and achieving the best reported results for the ISIC 2017 challenge dataset. Additionally, CAD systems have paid little attention to a critical criterion in skin lesion diagnosis: the lesion's evolution. This requires operating on two or more images of the same lesion, captured at different times but with a comparable scale, orientation and point of view; in other words, an image registration process should first be performed. We also propose in this work an image registration approach that outperforms top image registration techniques. Combined with the proposed lesion segmentation algorithm, this allows for the accurate extraction of features to assess the evolution of the lesion. We present a case-study with the lesion-size feature, paving the road for the development of automatic systems to easily evaluate skin lesion evolution.

A first step required to allow video indexing and retrieval of visual data is to perform a temporal segmentation, that is, to find the location of camera-shot transitions, which can be either abrupt (i.e., cuts) or gradual (e.g., fades, dissolves, wipes). After a critical review of most approaches seeking to solve this problem, we propose a unified detection model (both for abrupt and all types of gradual transitions), as well as an implementation whose results improve upon those of all the inspected reports. The innovation of the approach presented here is centered on mapping the space of inter-frame distances onto a new space of decision better suited to achieving a sequence-independent thresholding. This mapping aims to consider frame ordering information within the thresholding process; it is based on the parametric modeling of the patterns that transitions generate on the distances' output. As opposed to most reviewed works, our results are detailed over a large and representative sample of more than 1500 cuts and 250 gradual transitions, which make up a significant part (200 min) of the MPEG-7 testing material; this ensures a high degree of confidence in the validity of our approach.

In the current worldwide situation, pedestrian detection has reemerged as a pivotal tool for intelligent video-based systems aiming to solve tasks such as pedestrian tracking, social distancing monitoring or pedestrian mass counting. Pedestrian detection methods, even the top performing ones, are highly sensitive to occlusions among pedestrians, which dramatically degrades their performance in crowded scenarios. The generalization of multi-camera setups permits to better confront occlusions by combining information from different viewpoints. In this paper, we present a multi-camera approach to globally combine pedestrian detections leveraging automatically extracted scene context. Contrarily to the majority of the methods of the state-of-the-art, the proposed approach is scene-agnostic, not requiring a tailored adaptation to the target scenario–e.g., via fine-tuning. This noteworthy attribute does not require ad hoc training with labeled data, expediting the deployment of the proposed method in real-world situations. Context information, obtained via semantic segmentation, is used (1) to automatically generate a common area of interest for the scene and all the cameras, avoiding the usual need of manually defining it, and (2) to obtain detections for each camera by solving a global optimization problem that maximizes coherence of detections both in each 2D image and in the 3D scene. This process yields tightly fitted bounding boxes that circumvent occlusions or miss detections. The experimental results on five publicly available datasets show that the proposed approach outperforms state-of-the-art multi-camera pedestrian detectors, even some specifically trained on the target scenario, signifying the versatility and robustness of the proposed method without requiring ad hoc annotations nor human-guided configuration.

This paper describes the scientific achievements of a collaboration between a research group and the waste management division of a company. While these results might be the basis for several practical or commercial developments, we here focus on a novel scientific contribution: a methodology to automatically generate geo-located waste container maps. It is based on the use of Computer Vision algorithms to detect waste containers and identify their geographic location and dimensions. Algorithms analyze a video sequence and provide an automatic discrimination between images with and without containers. More precisely, two state-of-the-art object detectors based on deep learning techniques have been selected for testing, according to their performance and to their adaptability to an on-board real-time environment: EfficientDet and YOLOv5. Experimental results indicate that the proposed visual model for waste container detection is able to effectively operate with consistent performance disregarding the container type (organic waste, plastic, glass and paper recycling,…) and the city layout, which has been assessed by evaluating it on eleven different Spanish cities that vary in terms of size, climate, urban layout and containers' appearance.