Maciej Stefańczyk

Multimodal segmentation of dense depth maps and associated color information

An integrated segmentation approach for color images and depth maps is proposed. The 3D pointclouds are characterized by normal vectors and then grouped into planar, concave or convex faces. The empty regions in the depth map are filled by segments of the associated color image. In the experimental part two types of depth maps are analysed: generated by the MS-Kinect sensor or by a stereo-pair of cameras.

WUT Visual Perception Dataset: A Dataset for Registration and Recognition of Objects

Modern robots are typically equipped with many sensors with different modalities, e.g. RGB cameras, Time-of-Flight cameras or RGB-D sensors. Thus development of universal, modality-independent algorithms require appropriate datasets and benchmarks. In this paper we present WUT Visual Perception Dataset, consisting of five datasets, captured with different sensors with the goal of development, comparison and evaluation of algorithms for automatic object model registration and object recognition.

Control system of a service robot's active head exemplified on visual servoing

The article presents approach to active head control systems development. It describes the system structure including motor drivers with low level position control, as well as high level control mechanisms running on a computer. The formal specification of this system is based on the agent approach. Finally, an exemplary implementation of the active head control system for visual servoing application is presented. Experiments conducted on the system confirm the adequacy of the approach.

Reconfigurable control architecture for exploratory robots

Robots have to perform diverse and complex tasks. To face the limitations of the computational capabilities of robot on-board control computer, it is required to split the control systems between the robot embedded and the cloud computational resources. This paper presents a reconfigurable control architecture for a robot designed to meet this requirement. The embedded computer hosts a core agent, which provides the task-independent robot capabilities. The task dependent part called the dynamic agent is loaded from the cloud when required. The two mentioned agents execute the task, additionally utilising the capabilities of the cloud. Once the task is finished the dynamic agent is destroyed and the core agent awaits new user demands, upon which it reacts by downloading a new dynamic agent. Thus a reconfigurable system results, limited only by the resources provided by the cloud. The system is presented on an example of a humanoid robot exploring a home environment in search for hazards resulting from negligence of people suffering from mild dementia.

Utilization of Colour in ICP-based Point Cloud Registration

Advent of RGB-D sensors fostered the progress of computer vision algorithms spanning from object recognition, object and scene modelling to human activity recognition. This paper presents a new flavour of ICP algorithm developed for the purpose of pair-wise registration of colour point clouds generated from RGB-D images. After a brief introduction to the registration problem, we analyze the ICP algorithm and survey its different flavours in order to indicate potential methods of injecting of colour into it. Our consideration led to a solution, which we validate experimentally on colour point clouds from the publicly available dataset.

Efficient generation of 3D surfel maps using RGB–D sensors

Abstract The article focuses on the problem of building dense 3D occupancy maps using commercial RGB-D sensors and the SLAM approach. In particular, it addresses the problem of 3D map representations, which must be able both to store millions of points and to offer efficient update mechanisms. The proposed solution consists of two such key elements, visual odometry and surfel-based mapping, but it contains substantial improvements: storing the surfel maps in octree form and utilizing a frustum culling-based method to accelerate the map update step. The performed experiments verify the usefulness and efficiency of the developed system.

Localization of Essential Door Features for Mobile Manipulation

In this paper, we present a control system of a robot actively detecting and labeling door features, such as handle, lock or door plate. System is described using agent-based approach, with clear division into control, effector and receptor subsystems, with most important parts deeply explained. Presented approach supplements gap between door detection and opening algorithms, giving a robot the ability to take more actions, like recognizing room number or opening door lock.

Acquisition of RGB-D images: methods

The two-part article is devoted to sensors enabling the acquisition of depth information from the environment. The following, first part concentrates on three main methods of depth measurement: stereovision, structured light and time of flight (ToF). Along with the principle of operation of each of the method we also deliberate on their properties, analyse the complexity of required computations and present potential applications.

Handling of Asynchronous Data Flow in Robot Perception Subsystems

Robot perception subsystems typically form complex networks, with boxes representing computations and arrows presenting the exchanged data. Taking into account that data acquired from robot sensors may arrive with different frequencies, as well as that computations may by performed on different processor cores, a problem of handling of asynchronous data flows appears. Hence appropriate tools facilitating the implementation are highly demanded. In this article we propose a solution to the aforementioned problem, enabling the activation of a conditional behaviour of a given computational block, depending on the presence of data in its input buffers. Theoretical considerations led to the implementation of these mechanisms in a component-oriented framework for development of robot perception subsystems: DisCODe. Operation of the solution was verified on an exemplary perception subsystem using RGB-D camera.

Basic 3D Solid Recognition in RGB-D Images

The paper deals with the problem of recognition of 3D objects for the purpose of their subsequent grasping and manipulation by a two-handed robot. We describe the idea of a general framework for object recognition rooted in the compositional model of the world. This approach threats complex objects as entities constructed of simpler, elementary ones, termed solids. In particular, we focus on recognition of two types of such solids: cuboids and generalized cones. We present details of their operation, starting from the low-level processing of RGB-D images and ending with the generation of hypotheses regarding the presence and parameters of those types of solids.