Gangolf Hirtz

Bio: Gangolf Hirtz is an academic researcher from Chemnitz University of Technology. The author has contributed to research in topics: Orthogonal frequency-division multiplexing & Object detection. The author has an hindex of 8, co-authored 67 publications receiving 253 citations. Previous affiliations of Gangolf Hirtz include DST Systems & Information Technology University.

Automated real-time surveillance for ambient assisted living using an omnidirectional camera

Abstract: In this paper an automated video based real-time surveillance system is presented. This system is based on an omnidirectional camera and a multiple object tracking technique for applications in the field of AAL (Ambient Assisted Living). This system is able to monitor a complete room with a single camera and, in addition, to track the people entering and leaving this room. The software was implemented for an embedded platform which acts as a smart sensor.

Gait Parameter Estimation of Elderly People using 3D Human Pose Estimation in Early Detection of Dementia

Abstract: Early detection of dementia is becoming increasingly important as it plays a crucial role in handling the patients and offering better treatment. Many of the recent studies concluded a tight relationship between dementia and gait disorders. For this purpose, identification of gait abnormalities is key factor. Novel technologies provide many options such as wearable and non-wearable approaches for analysis of gait. As the occurrence of dementia is more prominent in elderly people, wearable technology is considered out of scope for this work. The gait data of several elderly people over 80 years is acquired over certain intervals during the scope of the project. The elderly people are classified into three study groups namely cognitively healthy individuals (CHI), subjectively cognitively impaired persons (SCI) and possible mildly cognitively impaired persons due to inconclusive test results (pMCI) based on their cognitive status. The gait data is acquired using Kinect sensor. The acquired data consists of both RGB image sequences and depth data of the test persons. 3D human pose estimation is performed on this gait data and gait analysis is done. The transformations in the gait cycles are observed and the health condition of the individual is analyzed. From the analysis, the patterns in the gait abnormalities are correlated with the above-mentioned classification and are used in the detection of dementia in advance. The obtained results look promising and further analysis of gait parameters is under progress.

A fast approach for omnidirectional surveillance with multiple virtual perspective views

Abstract: In recent years, video surveillance combined with computer vision algorithms like object detection, tracking or automated behaviour analysis has become an important research topic. However, most of these systems are depending on either fixed or remotely controlled narrow angle cameras. When using the former, the area of coverage is extremely limited, while utilizing the latter leads to high failure rates and troubles in camera calibration. In this paper, a method of extracting multiple perspective views from a single omnidirectional image for realtime environments is proposed. An example application of a ceiling-mounted camera setup is used to show the functional principle. Furthermore a performance improvement strategy is both presented and evaluated.

Signal fusion based on intensity and motion variations for remote heart rate determination

Abstract: In recent years, e-rehabilitation has become an emerging topic, firstly because of an increasing demand, secondly because of improved sensor systems and higher computational performance. Furthermore, due to the lack of therapists in Germany, an adequate supervision of the therapy is often impossible. A tracking of physiological parameters, such as the heart rate, can contribute to an improved evaluation of training exercise efficiency. In this work, we present several methods to remotely determine the heart rate with an RGB camera. To achieve this, we propose two fusion methods , which combine intensity-based and motion-based methods. The fusion method with the highest performance is based on the measurement of intensity and motion variations. This new approach outperforms extant single methods.

Sensor-based systems for early detection of dementia (SENDA): a study protocol for a prospective cohort sequential study

Abstract: Dementia and cognitive decline are serious social and economic burdens. An increase in the population of older people, as well as longer lifespans mean that numbers of dementia cases are exponentially rising. Neuropathological changes associated with dementia are thought to appear before the clinical manifestation of cognitive symptoms, i.e., memory impairments. Further, some older adults (OA) experience cognitive decline before it can be objectively diagnosed. For optimal care of these patients, it is necessary to detect cognitive decline and dementia at an early stage. In this vein, motor, sensory, and neurophysiological declines could be promising factors if found to be present before the onset of cognitive impairment. Hence, the objective of the SENDA study is to develop a multi-dimensional sensor-based instrument that allows early detection of cognitive decline or dementia in OA with the help of cognitive, sensory, motor, and neurophysiological parameters before its clinical manifestation. In the cohort sequential study, participants are assigned to one of three study groups depending on their cognitive status: 1. cognitively healthy individuals (CHI), 2. subjectively cognitively impaired persons (SCI), or 3. (possible) mildly cognitively impaired persons (pMCI, MCI). All groups take part in the same cognitive (e.g., executive function tests), motor (e.g., gait analyses, balance tests), sensory (e.g., vibration perception threshold test, proprioception tests), and neurophysiological (e.g., electroencephalograms) measurements. Depending on the time at which participants are included into the study, all measurements are repeated up to four times in intervals of 8 months within 3 years to identify associations with cognitive changes over time. This study aims to detect possible motor, sensory, neurophysiological, and cognitive predictors to develop an early screening tool for dementia and its pre-stages in OA. Thus, affected persons could receive optimal health care at an earlier time point to maintain their health resources. The study is ongoing. The recruitment of participants will be continued until May 2020.

Learning without Forgetting

Abstract: When building a unified vision system or gradually adding new capabilities to a system, the usual assumption is that training data for all tasks is always available. However, as the number of tasks grows, storing and retraining on such data becomes infeasible. A new problem arises where we add new capabilities to a Convolutional Neural Network (CNN), but the training data for its existing capabilities are unavailable. We propose our Learning without Forgetting method, which uses only new task data to train the network while preserving the original capabilities. Our method performs favorably compared to commonly used feature extraction and fine-tuning adaption techniques and performs similarly to multitask learning that uses original task data we assume unavailable. A more surprising observation is that Learning without Forgetting may be able to replace fine-tuning with similar old and new task datasets for improved new task performance.

Gliding Vertex on the Horizontal Bounding Box for Multi-Oriented Object Detection

Abstract: Object detection has recently experienced substantial progress. Yet, the widely adopted horizontal bounding box representation is not appropriate for ubiquitous oriented objects such as objects in aerial images and scene texts. In this paper, we propose a simple yet effective framework to detect multi-oriented objects. Instead of directly regressing the four vertices, we glide the vertex of the horizontal bounding box on each corresponding side to accurately describe a multi-oriented object. Specifically, We regress four length ratios characterizing the relative gliding offset on each corresponding side. This may facilitate the offset learning and avoid the confusion issue of sequential label points for oriented objects. To further remedy the confusion issue for nearly horizontal objects, we also introduce an obliquity factor based on area ratio between the object and its horizontal bounding box, guiding the selection of horizontal or oriented detection for each object. We add these five extra target variables to the regression head of faster R-CNN, which requires ignorable extra computation time. Extensive experimental results demonstrate that without bells and whistles, the proposed method achieves superior performances on multiple multi-oriented object detection benchmarks including object detection in aerial images, scene text detection, pedestrian detection in fisheye images.