Refine
Document Type
- Conference proceeding (8)
- Journal article (1)
Language
- English (9)
Has full text
- yes (9)
Is part of the Bibliography
- yes (9)
Institute
- Technik (7)
- Informatik (2)
Publisher
- Springer (3)
- IEEE (2)
- Cornell University (1)
- International Academy Publishing (1)
- SciTePress (1)
- Springer International Publishing (1)
This paper presents a generic method to enhance performance and incorporate temporal information for cardiorespiratory-based sleep stage classification with a limited feature set and limited data. The classification algorithm relies on random forests and a feature set extracted from long-time home monitoring for sleep analysis. Employing temporal feature stacking, the system could be significantly improved in terms of Cohen’s κ and accuracy. The detection performance could be improved for three classes of sleep stages (Wake, REM, Non-REM sleep), four classes (Wake, Non-REM-Light sleep, Non-REM Deep sleep, REM sleep), and five classes (Wake, N1, N2, N3/4, REM sleep) from a κ of 0.44 to 0.58, 0.33 to 0.51, and 0.28 to 0.44 respectively by stacking features before and after the epoch to be classified. Further analysis was done for the optimal length and combination method for this stacking approach. Overall, three methods and a variable duration between 30 s and 30 min have been analyzed. Overnight recordings of 36 healthy subjects from the Interdisciplinary Center for Sleep Medicine at Charité-Universitätsmedizin Berlin and Leave-One-Out-Cross-Validation on a patient-level have been used to validate the method.
For collision and obstacle avoidance as well as trajectory planning, robots usually generate and use a simple 2D costmap without any semantic information about the detected obstacles. Thus a robot’s path planning will simply adhere to an arbitrarily large safety margin around obstacles. A more optimal approach is to adjust this safety margin according to the class of an obstacle. For class prediction, an image processing convolutional neural network can be trained. One of the problems in the development and training of any neural network is the creation of a training dataset. The first part of this work describes methods and free open source software, allowing a fast generation of annotated datasets. Our pipeline can be applied to various objects and environment settings and is extremely easy to use to anyone for synthesising training data from 3D source data. We create a fully synthetic industrial environment dataset with 10 k physically-based rendered images and annotations. Our da taset and sources are publicly available at https://github.com/LJMP/synthetic-industrial-dataset. Subsequently, we train a convolutional neural network with our dataset for costmap safety class prediction. We analyse different class combinations and show that learning the safety classes end-to-end directly with a small dataset, instead of using a class lookup table, improves the quantity and precision of the predictions.
Efficient and robust 3D object reconstruction based on monocular SLAM and CNN semantic segmentation
(2019)
Various applications implement slam technology, especially in the field of robot navigation. We show the advantage of slam technology for independent 3d object reconstruction. To receive a point cloud of every object of interest void of its environment, we leverage deep learning. We utilize recent cnn deep learning research for accurate semantic segmentation of objects. In this work, we propose two fusion methods for cnn-based semantic segmentation and slam for the 3d reconstruction of objects of interest in order to obtain a more robustness and efficiency. As a major novelty, we introduce a cnn-based masking to focus slam only on feature points belonging to every single object. Noisy, complex or even non-rigid features in the background are filtered out, improving the estimation of the camera pose and the 3d point cloud of each object. Our experiments are constrained to the reconstruction of industrial objects. We present an analysis of the accuracy and performance of each method and compare the two methods describing their pros and cons.
Recognition of sleep and wake states is one of the relevant parts of sleep analysis. Performing this measurement in a contactless way increases comfort for the users. We present an approach evaluating only movement and respiratory signals to achieve recognition, which can be measured non-obtrusively. The algorithm is based on multinomial logistic regression and analyses features extracted out of mentioned above signals. These features were identified and developed after performing fundamental research on characteristics of vital signals during sleep. The achieved accuracy of 87% with the Cohen’s kappa of 0.40 demonstrates the appropriateness of a chosen method and encourages continuing research on this topic.
Facial beauty prediction (FBP) aims to develop a machine that automatically makes facial attractiveness assessment. In the past those results were highly correlated with human ratings, therefore also with their bias in annotating. As artificial intelligence can have racist and discriminatory tendencies, the cause of skews in the data must be identified. Development of training data and AI algorithms that are robust against biased information is a new challenge for scientists. As aesthetic judgement usually is biased, we want to take it one step further and propose an Unbiased Convolutional Neural Network for FBP. While it is possible to create network models that can rate attractiveness of faces on a high level, from an ethical point of view, it is equally important to make sure the model is unbiased. In this work, we introduce AestheticNet, a state-of-the-art attractiveness prediction network, which significantly outperforms competitors with a Pearson Correlation of 0.9601. Additionally, we propose a new approach for generating a bias-free CNN to improve fairness in machine learning.
For autonomously driving cars and intelligent vehicles it is crucial to understand the scene context including objects in the surrounding. A fundamental technique accomplishing this is scene labeling. That is, assigning a semantic class to each pixel in a scene image. This task is commonly tackled quite well by fully convolutional neural networks (FCN). Crucial factors are a small model size and a low execution time. This work presents the first method that exploits depth cues together with confidence estimates in a CNN. To this end, novel experimentally grounded network architecture is proposed to perform robust scene labeling that does not require costly preprocessing like CRFs or LSTMs as commonly used in related work. The effectiveness of this approach is demonstrated in an extensive evaluation on a challenging real-world dataset. The new architecture is highly optimized for high accuracy and low execution time.
SLAM systems are mainly applied for robot navigation while research on feasibility for motion planning with SLAM for tasks like bin-picking, is scarce. Accurate 3D reconstruction of objects and environments is important for planning motion and computing optimal gripper pose to grasp objects. In this work, we propose the methods to analyze the accuracy of a 3D environment reconstructed using a LSD-SLAM system with a monocular camera mounted onto the gripper of a collaborative robot. We discuss and propose a solution to the pose space conversion problem. Finally, we present several criteria to analyze the 3D reconstruction accuracy. These could be used as guidelines to improve the accuracy of 3D reconstructions with monocular LSD-SLAM and other SLAM based solutions.
In the last 20 years there have been major advances in autonomous robotics. In IoT (Industry 4.0), mobile robots require more intuitive interaction possibilities with humans in order to expand its field of applications. This paper describes a user-friendly setup, which enables a person to lead the robot in an unknown environment. The environment has to be perceived by means of sensory input. For realizing a cost and resource efficient Follow Me application we use a single monocular camera as low-cost sensor. For efficient scaling of our Simultaneous Localization and Mapping (SLAM) algorithm, we integrate an inertial measurement unit (IMU) sensor. With the camera input we detect and track a person. We propose combining state of the art deep learning with Convolutional Neural Network (CNN) and SLAM algorithms functionality on the same input camera image. Based on the output robot navigation is possible. This work presents the specification, workflow for an efficient development of the Follow Me application. Our application’s delivered point clouds are also used for surface construction. For demonstration, we use our platform SCITOS G5 equipped with the afore mentioned sensors. Preliminary tests show the system works robustly in the wild.
This paper presents a novel multi-modal CNN architecture that exploits complementary input cues in addition to sole color information. The joint model implements a mid-level fusion that allows the network to exploit cross modal interdependencies already on a medium feature-level. The benefit of the presented architecture is shown for the RGB-D image understanding task. So far, state-of-the-art RGB-D CNNs have used network weights trained on color data. In contrast, a superior initialization scheme is proposed to pre-train the depth branch of the multi-modal CNN independently. In an end-to-end training the network parameters are optimized jointly using the challenging Cityscapes dataset. In thorough experiments, the effectiveness of the proposed model is shown. Both, the RGB GoogLeNet and further RGB-D baselines are outperformed with a significant margin on two different tasks: semantic segmentation and object detection. For the latter, this paper shows how to extract object level groundtruth from the instance level annotations in Cityscapes in order to train a powerful object detector.
