Guillermo Cámara-Chávez

ECG-based heartbeat classification for arrhythmia detection

An electrocardiogram (ECG) measures the electric activity of the heart and has been widely used for detecting heart diseases due to its simplicity and non-invasive nature. By analyzing the electrical signal of each heartbeat, i.e., the combination of action impulse waveforms produced by different specialized cardiac tissues found in the heart, it is possible to detect some of its abnormalities. In the last decades, several works were developed to produce automatic ECG-based heartbeat classification methods. In this work, we survey the current state-of-the-art methods of ECG-based automated abnormalities heartbeat classification by presenting the ECG signal preprocessing, the heartbeat segmentation techniques, the feature description methods and the learning algorithms used. In addition, we describe some of the databases used for evaluation of methods indicated by a well-known standard developed by the Association for the Advancement of Medical Instrumentation (AAMI) and described in ANSI/AAMI EC57:1998/(R)2008 (ANSI/AAMI, 2008). Finally, we discuss limitations and drawbacks of the methods in the literature presenting concluding remarks and future challenges, and also we propose an evaluation process workflow to guide authors in future works.

Detection of Groups of People in Surveillance Videos Based on Spatio-Temporal Clues

Video surveillance has been widely employed in our society in the past years. In this context, humans play an important role and are the major players since they are responsible for changing the state of the scene through actions and activities. Therefore, the design of automatic methods to understand human behavior and recognize activities are important to determine which subjects are involved in an activity of interest. The computer vision research area has contributed vastly for the development of methods related to detection, tracking and recognition of humans. However, there is still a lack of methods able to recognize higher level activities (e.g., interaction among people that might be involved in an illegal activity). The first step to be successful in this enterprise is to detect and locate groups of people in the scene, which is essential to make inferences regarding interactions among persons. Aiming at such direction, this paper presents a group detection approach that combines motion and spatial information with low-level descriptors to be robust to situations such as partial occlusions. The experimental results obtained using the PETS 2009 and the BEHAVE datasets demonstrate that the proposed combination indeed achieves higher accuracies, indicating a promising direction for future research.

An Adaptive Vehicle License Plate Detection at Higher Matching Degree

In this paper, a novel approach for vehicle license plate detection that improves in both efficiency and quality over the common multiscale search method is proposed. The detection efficiency is improved by employing the result of a single scale sliding window search as a promising guess of the license plate location. The quality is assured by locally refining the initial detection in multiple scales. The main benefit of our method is that we have reached a more precise detection with the analysis of 20 times fewer detection windows with high reliability (96% recall and 70% precision). We also compared our method with an edge-based hybrid approach.

Harris-SIFT Descriptor for Video Event Detection Based on a Machine Learning Approach

Video data is becoming increasingly important in many commercial and scientific areas with the advent of applications such as digital broadcasting, video-conferencing and multimedia processing tools, and with the development of the hardware and communications infrastructure necessary to support visual applications. The objective of this work is to propose a method for event detection in a video stream. We combine Harris-SIFT descriptor with motion information in order to detect human actions in video. We tested our method in KTH database and compared it to space-time interest points (STIP) descriptor. The results obtained achieved similar results to the STIP method.

Building semantic understanding beyond deep learning from sound and vision

Deep learning-based models have recently been widely successful at outperforming traditional approaches in several computer vision applications such as image classification, object recognition and action recognition. However, those models are not naturally designed to learn structural information that can be important to tasks such as human pose estimation and structured semantic interpretation of video events. In this paper, we demonstrate how to build structured semantic understanding of audio-video events by reasoning on multiple-label decisions of deep visual models and auditory models using Grenanders structures for imposing semantic consistency. The proposed structured model does not require joint training of the structural semantic dependencies and deep models. Instead they are independent components linked by Grenanders structures. Furthermore, we exploited Grenanders structures as a means to facilitate and enrich the model with fusion of multimodal sensory data; in particular, auditory features with visual features. Overall, we observed improvements in the quality of semantic interpretations using deep models and auditory features in combination with Grenanders structures, reflecting as numerical improvements of up to 11.5% and 12.3% in precision and recall, respectively.

GPUs and Multicore CPUs Implementations of a Static Video Summarization

The fast evolution of digital media, in special digital videos, has created an exponential growth of data, increasing the storage and transmission cost and the video content retrieve information complexity. Video summarization has been proposed to circumvent some of these issues and also serves as a pre-processing step in many video applications. In this paper, a static video summarization algorithm is studied and in order to reduce its high execution time, parallelizations using Graphics Processor Units (GPUs) and multicore CPUs are proposed. We also explore a hybrid approach combining both hardware to maximize the performance. The experiments were performed using 120 videos varying frame resolution and video length and the results showed that the hybrid and the multicore CPUs versions reached the best executions times, achieving 4× speedup in average.

Real-Time Brand Logo Recognition

The increasing popularity of Social Networks makes change the way people interact. These interactions produce a huge amount of data and it opens the door to new strategies and marketing analysis. According to Instagram (https://instagram.com/press/) and Tumblr (https://www.tumblr.com/press), an average of 80 and 59 million photos respectively are published every day, and those pictures contain several implicit or explicit brand logos. The analysis and detection of logos in natural images can provide information about how widespread is a brand. In this paper, we propose a real-time brand logo recognition system, that outperforms all other state-of-the-art methods for the challenging FlickrLogos-32 dataset. We experimented with 5 different approaches, all based on the Single Shot MultiBox Detector (SSD). Our best results were achieved with the SSD 512 pretrained, where we outperform by 2.5% of F-score and by 7.4% of recall the best results on this dataset. Besides the higher accuracy, this approach is also relatively fast and can process with a single Nvidia Titan X 19 images per second.

Fusion of Deep Learning Descriptors for Gesture Recognition.

In this paper, we propose an approach for dynamic hand gesture recognition, which exploits depth and skeleton joint data captured by Kinect™ sensor. Also, we select the most relevant points in the hand trajectory with our proposed method to extract keyframes, reducing the processing time in a video. In addition, this approach combines pose and motion information of a dynamic hand gesture, taking advantage of the transfer learning property of CNNs. First, we use the optical flow method to generate a flow image for each keyframe, next we extract the pose and motion information using two pre-trained CNNs: a CNN-flow for flow-images and a CNN-pose for depth-images. Finally, we analyze different schemes to fusion both informations in order to achieve the best method. The proposed approach was evaluated in different datasets, achieving promising results compared to other methods, outperforming state-of-the-art methods.