Due to the short and noisy nature of Twitter microposts, detecting named entities is often a cumbersome task. As part of the ACL2015 Named Entity Recognition (NER) shared task, we present a semisupervised system that detects 10 types of named entities. To that end, we leverage 400 million Twitter microposts to generate powerful word embeddings as input features and use a neural network to execute the classification. To further boost the performance, we employ dropout to train the network and leaky Rectified Linear Units (ReLUs). Our system achieved the fourth position in the final ranking, without using any kind of hand-crafted features such as lexical features or gazetteers.

No abstract

Many smart home applications rely on indoor human activity recognition. This challenge is currently primarily tackled by employing video camera sensors. However, the use of such sensors is characterized by fundamental technical deficiencies in an indoor environment, often also resulting in a breach of privacy. In contrast, a radar sensor resolves most of these flaws and maintains privacy in particular. In this paper, we investigate a novel approach towards automatic indoor human activity recognition, feeding high-dimensional radar and video camera sensor data into several deep neural networks. Furthermore, we explore the efficacy of sensor fusion to provide a solution in less than ideal circumstances. We validate our approach on two newly constructed and published data sets that consist of 2347 and 1505 samples distributed over six different types of gestures and events, respectively. From our analysis, we can conclude that, when considering a radar sensor, it is optimal to make use of a three-dimensional convolutional neural network that takes as input sequential range-Doppler maps. This model achieves 12.22% and 2.97% error rate on the gestures and the events data set, respectively. A pre-trained residual network is employed to deal with the video camera sensor data and obtains 1.67% and 3.00% error rate on the same data sets. We show that there exists a clear benefit in com-Baptist Vandersmissen

In this paper, we tackle the task of multi-target tracking of humans in an indoor setting using a low power 77 GHz MIMO CMOS radar. A drawback of such a highresolution and low-power device is the higher sensitivity to noise, which makes the analysis of signals more challenging. Therefore, a pipeline is proposed to address both pre-processing of the radar signal and multi-target tracking. In the pre-processing phase, we focus on handling the low Signal-to-Noise Ratio (SNR) and eliminating so-called ghost targets. The tracking method we propose is based on Markov Chain Monte Carlo Data Association (MCMCDA), thus taking a combinatorial approach towards the task of tracking. The pipeline is tested on a number of real-world scenarios and shows promising results, overcoming the significant amount of noise associated with embedded radar devices.

No abstract

Bone age is an essential measure of skeletal maturity in children with growth disorders. It is typically assessed by a trained physician using radiographs of the hand and a reference model. However, it has been described that the reference models leave room for interpretation leading to a large inter-observer and intra-observer variation. In this work, we explore a novel method for automated bone age assessment to assist physicians with their estimation. It consists of a powerful combination of deep learning and Gaussian process regression. Using this combination, sensitivity of the deep learning model to rotations and flips of the input images can be exploited to increase overall predictive performance compared to only using the deep learning network. We validate our approach retrospectively on a set of 12611 radiographs of patients between 0 and 19 years of age.

Along with substantial advances in the area of image processing and, consequently, video-based surveillance systems, concerns about preserving the privacy of people have also deepened. Therefore, replacing conventional video cameras in surveillance systems with less-intrusive and yet effective alternatives, such as micro-wave radars, is of high interest. The aim of this work is to explore the application of Reservoir Computing Networks (RCNs) to the problem of identifying a limited number of people in an indoor environment, leveraging gait information captured by micro-wave radar measurements. These measurements are done using a commercial low-power linear frequency-modulated continuous-wave (FMCW) radar.Besides the low quality of the outputs of such a radar sensor, walking spontaneously as opposed to controlled situations adds another level of complexity to the targeted use case. In this context, RCNs are interesting tools, given that they have shown a high effectiveness in capturing temporal information and handling noise, while at the same time being easy to setup and train. Using Micro-Doppler features as inputs, we follow a structured procedure towards optimizing the parameters of our RCN-based approach, showing that RCNs have a great potential in processing the noisy features provided by a low-power radar.

Video cameras are arguably the world's most used sensors for surveillance systems. They give a highly detailed representation of a situation that is easily interpreted by both humans and computers. However, these representations can lose part of their representational value when being recorded in less than ideal circumstances. Bad weather conditions, low-light illumination or concealing objects can make the representation more opaque. A radar sensor is a potential solution for these situations, since it is unaffected by the light intensity and can sense through most concealing objects. In this paper, we investigate the performance of a structured inference network on data of a low-power radar device. A structured inference network applies automated feature extraction by creating a latent space out of which the observations can be reconstructed. A classification model can then be trained on this latent space. This methodology allows us to perform experiments for both person identification and action recognition, resulting in competitive error rates ranging from 0% to 6.5% for actions recognition and 10% to 12% for person identification. Furthermore, the possibility of a radar sensor being used as a complement to a camera sensor is investigated.