Carsten Stahlschmidt

Applications for a people detection and tracking algorithm using a time-of-flight camera

This paper outlines a method and applications for detection and tracking of people in depth images, acquired with a low-resolution Time-of-Flight (ToF) camera. This depth sensor is placed perpendicular to the ground in order to provide distance information from a top-view position. Usage of intrinsic and extrinsic camera parameters allows estimation of a ground plane and comparison to the measured distances of the ToF sensor in every pixel. Differences to the expected ground plane define foreground information, that is subsequently combined to associated regions. These regions of interest (ROI) are analyzed to distinguish persons from other objects by using a matched filter that is applied the height segmented depth information of each of these regions. The proposed method separates crowds into individuals and facilitates a multi-object tracking system based on Kalman filtering. Furthermore, we present several applications for the proposed method. Experiments with different crowding situations - from very low to very high density - and different heights of camera placements have proven the applicability and practicability of the system.

Descending step classification using time-of-flight sensor data

This paper proposes a method to analyse human-made environments regarding the existence of descending stairs and steps to assists visually impaired and furthermore disabled people, that are not able to use traditional supports like blind canes. Those people are heavily limited in their daily lives, since wrong decisions caused by the lack of information can easily lead to accidents. We use depth data acquired with a low-resolution Time-of-Flight (ToF) camera to perceive the scene in front a mobile vehicle (rollator) to provide the user with detailed information about potentially hazardous situations. Experiments with affected persons have shown the ability of the system to help them understand the environment and, in particular, avoid falls from descending stairs.

People Detection and Tracking from a Top-View Position Using a Time-of-Flight Camera

This paper outlines a method for the detection and tracking of people in depth images, acquired with a low-resolution Time-of-Flight (ToF) camera. This depth sensor is placed perpendicular to the ground in order to provide distance information from a top-view position.

RSSI-Based Real-Time Indoor Positioning Using ZigBee Technology for Security Applications

Localization in indoor environments is an important aspect with regard to mobile security applications. Because here, the global positioning system (GPS) is not available or very imprecise, other positioning systems are required. For that matter wireless sensor networks provide two common approaches based on received signal strength indicators (RSSI). The first one uses fingerprints and the second is based on trilateration. Because fingerprinting needs a lot of training and (re-)calibration, this paper presents a new indoor positioning system based on RSSIs and trilateration using ZigBee technology. Since RSSI measurements are very susceptible to noise, the gathered RSSIs have to be preprocessed before they can be used for position calculations. For this reason, the RSSIs were averaged using time-dependent weights and smoothed over time so that outliers and old RSSIs can be eliminated. The presented indoor positioning system was verified by experiments.

Classification of ascending steps and stairs using Time-of-Flight sensor data

This paper proposes a method to analyse human-made environments in order to verify the presence of ascending steps or stairs. Our system is intended to assist visually impaired people by providing acoustic information about the scene in front of a low-resolution Time-of-Flight (ToF) camera that is fixed to a mobile vehicle (rollator). Detailed instructions to the user about potentially hazardous situations are provided. This paper in particular deals with a fast approach on classification of ascending steps in 3D point clouds. This method is part of a system that aims on enhancing visually impaired persons understand the environment and help prevent collisions.

Evaluation of pedestrian detection fusion and localization based on the idea of car-to-X communication

Pedestrian safety applications for urban environments, e.g. observation of intersections or crosswalks for pedestrians, are a growing research area. This paper is focused on the fusion of detected pedestrians from different viewing angles, especially from a surveillance system and from a moving car. Therefore, a combination of the available information inside of a generated world map, which is based on the global positioning system (GPS), is presented. The model for transforming detections with assumed uncertainties into the world map, and an attempt to combine separate independent single detections from different views to a unique one are results of the paper. Besides the combination of information, an evaluation of the detection accuracy of a surveillance system and a monocular object detection driver assistance system is presented. This evaluation discloses drawbacks as well as benefits of an information mapping from different viewing angles into a global world map.

Feature Recycling Cascaded SVM Classifier Based on Feature Selection of HOGs for Pedestrian Detection

Since to pedestrian detection in driver assistance as well as surveillance systems is a challenging task of the recent years this paper introduces a fast cascaded classifier based on linear and non-linear support vector machines (SVMs). To yield high and accurate detection rates, histogram of oriented gradients (HOGs) will be preselected by the fisher score. These features will be a basis for the training algorithm of the cascaded classifier. A non-maximum suppression algorithm will be used and evaluated in respect to reject HOG features which have a huge overlap in a joint image area. By variation of the non-maximum suppression parameter different numbers of preselected HOG features will be used to create the cascaded classifier. The different cascaded classifiers will be evaluated and compared between each other and in relation to the HOG procedure from Dalal and Triggs combined with a support vector machine.

Posture independent stair parameter estimation

This paper proposes a method to analyse humanmade environments in order to identify ascending steps and stairs and recognize their parameters. The system uses adjustable 3D environment geometries for the detection of steps and estimates tread depth and width, riser height and yaw orientation from the camera to staircase. Those parameters are needed by humanoid robots to ensure a safe traversal of staircases. The camera perceives the scene in front of a mobile vehicle with a system that is originally designed to provide visually impaired and furthermore disabled people with detailed information about potentially hazardous situations. Since most of the affected persons might not want to traverse staircases, this work is mostly suitable to be used for multifloor exploring humanoid robots. Experiments substantiate the ability to provide good parameters, independent from posture to staircases. This even works with relatively large distances to staircases (≥ 2m) and enables an early adaptation of navigation.

Automatic extrinsic camera parameters estimation for a mobile ToF camera application

Extrinsic camera parameters estimation is an important task for many assistance systems. The reconstruction of the image to world projection depends strongly on robust estimated camera parameters. Based on the used camera system, the extrinsic parameters estimation can be a complicated task, that maybe requires a lot of computing capacity. In this paper, an extrinsic camera parameters estimation procedure for a time of flight sensor, equipped on a rollator for impaired people, is presented. Effectiveness and robustness of the procedure are in the focus of the developed calibration procedure. The precision of image to world projection based on the estimated parameters is evaluated. The influence of the sensor noise regarding the measured diagonal values, e.g. depth values in the vicinity of the principal point of the camera sensor, to the ground plane is evaluated, too. Many different viewing angles as well as different height configurations of the system setup are used to evaluate the accuracy of the estimation procedure.

LDA Combined Depth Similarity and Gradient Features for Human Detection using a Time-of-Flight Sensor

Visual object detection is an important task for many research areas like driver assistance systems (DASs), industrial automation and various safety applications with human interaction. Since detection of pedestrians is a growing research area, different kinds of visual methods and sensors have been introduced to overcome this problem. This paper introduces new relational depth similarity features (RDSF) for the pedestrian detection using a Time-of-Flight (ToF) camera sensor. The new features are based on mean, variance, skewness and kurtosis values of local regions inside the depth image generated by the Time-of-Flight sensor. An evaluation between these new features, already existing relational depth similarity features using depth histograms of local regions and the well known histogram of oriented gradients (HOGs), which deliver very good results in the topic of pedestrian detection, will be presented. To incorporate more dimensional feature spaces, an existing AdaBoost algorithm, which uses linear discriminant analysis (LDA) for feature space reduction and new combination of already extracted features in the training procedure, will be presented too.