Abstract-We present a vision based multisensor that is designed for robot interaction with small, soft, and possibly fragile objects. The sensor consists of a rubber membrane, a rectangular frame on which the membrane is mounted and a CCD camera. The entire system is airtight. Based on the observed deformations of the membrane, we determine the contact area, the integral force acting on the membrane, the 3D force distribution over the membrane, and derive properties of the target object by monitoring the evolution of its deformation. We can distinguish between different types of materials, i.e., solid, soft, amorphous, and determine the speed and nature of their deformation. The sensitivity of the sensor can be adjusted by changing the volume of air within the rectangular frame. We achieved a small noise to signal ratio, which allows us to observe small integral forces in the range of 0.5 N to 2.5 N, with an average error of 0.04 N.

A novel method for quantitatively measuring social interactions on small temporal and spatial scales on the basis of interaction geometry (reduced to the parameters interpersonal distance and relative body orientation) with the help of infrared (IR) tracking is introduced. The method is intended to be used to establish a probabilistic classifier to identify existing social situations on the basis of measuring the two parameters for pairs of persons through a series of experiments. The classifier can then be used for characterizing the social context (as an evidence for or against established social situations) of users using sensors in mobile devices in view of useful future Mobile Social Networking services. A first experiment is conducted with the method, a number of standard classifiers including a Gaussian Mixture Model are trained and evaluated and the results are discussed.

No abstract

Lane change prediction of surrounding vehicles is a key building block of path planning. The focus has been on increasing the accuracy of prediction by posing it purely as a function estimation problem at the cost of model understandability. However, the efficacy of any lane change prediction model can be improved when both corner and failure cases are humanly understandable. We propose an attentionbased recurrent model to tackle both understandability and prediction quality. We also propose metrics which reflect the discomfort felt by the driver. We show encouraging results on a publicly available dataset and proprietary fleet data.

In this paper, we present a method for human full-body pose estimation from Time-of-Flight (ToF) camera images. Our approach consists of robustly detecting anatomical landmarks in the 3D data and fitting a skeleton body model using constrained inverse kinematics. Instead of relying on appearance-based features for interest point detection that can vary strongly with illumination and pose changes, we build upon a graph-based representation of the ToF depth data that allows us to measure geodesic distances between body parts. As these distances do not change with body movement, we are able to localize anatomical landmarks independent of pose. For differentiation of body parts that occlude each other, we employ motion information, obtained from the optical flow between subsequent ToF intensity images. We provide a qualitative and quantitative evaluation of our pose tracking method on ToF sequences containing movements of varying complexity.

In this paper, we propose a method for simultaneous human full-body pose tracking and activity recognition from time-of-flight (ToF) camera images. Simple and sparse depth cues are used together with a prior motion model that constrains the tracking problem. Our model consists of low-dimensional manifolds of feasible poses for multiple activities. A particle filter allows us to efficiently evaluate various pose hypotheses over different activities and to select one that is most consistent with the observed depth image cues. We relate poses in the manifold embeddings to full-body poses and to observable depth cues using non-linear regression mappings. Our method is able to robustly detect changes of activity and adapt accordingly. We evaluate our method on a dataset containing 10 activities for 10 persons and show that we can track full-body pose and classify performed activities with a high precision which is discussed in the paper.

Abstract. In this paper, a method is presented that allows reconstructing the full-body pose of a person in real-time, based on the limited input from a few wearable inertial sensors. Our method uses Gaussian Process Regression to learn the person-specific functional relationship between the sensor measurements and full-body pose. We generate training data by recording sample movements for different activities simultaneously using inertial sensors and an optical motion capture system. Since our approach is discriminative, pose prediction from sensor data is efficient and does not require manual initialization or iterative optimization in pose space. We also propose a SVM-based scheme to classify the activities based on inertial sensor data. An evaluation is performed on a dataset of movements, such as walking or golfing, performed by different actors. Our method is capable of reconstructing the full-body pose from as little as four inertial sensors with an average angular error of 4-6 degrees per joint, as shown in our experiments.

No abstract