Rafal Kapela

Real-time shape description system based on MPEG-7 descriptors

The paper presents a real-time shape description system based on MPEG-7 standard descriptors. The proposed system architecture is used for description in real-time image objects shape basing on features such as contour and occupied region. The proposed hardware architecture splits the computational burden into several processes where calculations of mentioned objects features are made simultaneously in order to improve systems speed. A novel approximation method for estimation of peaks height in the curvature scale space (CSS) representation of the shape as well as the artificial neural network (ANN) used in system development process for generalization of the data samples received from the approximation module. These methods make hardware realizations of main computational-consuming modules of the system more time efficient.

Real-time field sports scene classification using colour and frequency space decompositions

This paper presents a novel approach to recognize a scene presented in an image with specific application to scene classification in field sports video. We propose different variants of the algorithm ranging from bags of visual words to the simplified real-time implementation, that takes only the most important areas of similar colour into account. All the variants feature similar accuracy which is comparable to very well-known image indexing techniques like SIFT or HoGs. For the comparison purposes, we also developed a specific database which is now available online. The algorithm is suitable in scene recognition task thanks to changes in speed and robustness to the image resolution, thus, making it a good candidate in real-time video indexing systems. The procedure features high simplicity thanks to the fact that it is based on the very well-known Fourier transform.

Asphalt surfaced pavement cracks detection based on histograms of oriented gradients

Cracks are the most requiring type of pavement distresses to detect and classify automatically. Due to its nature are easily absorbed by other types of pavement surface damages. Moreover, the diversity of pavement surface makes the image detection system requiring efficient computer algorithms. The paper presents the solutions tested on surface distress data which were collected automatically using downward facing cameras placed orthogonally to road pavement axis. Presented results focus on the crack-type pavement distresses. The achieved accuracy of the transverse, longitudinal and meshing cracks recognition based on the initial dataset prepared especially for this system, show it has very good chances to work efficiently with large image datasets collected during the inspection car runs.

Real-time event classification in field sport videos

The paper presents a novel approach to real-time event detection in sports broadcasts. We present how the same underlying audio-visual feature extraction algorithm based on new global image descriptors is robust across a range of different sports alleviating the need to tailor it to a particular sport. In addition, we propose and evaluate three different classifiers in order to detect events using these features: a feed-forward neural network, an Elman neural network and a decision tree. Each is investigated and evaluated in terms of their usefulness for real-time event classification. We also propose a ground truth dataset together with an annotation technique for performance evaluation of each classifier useful to others interested in this problem. HighlightsArticle presents novel system for real-time video indexing for field sports videos based on the new video description technique.Up to date there is no system that covers so many different sports genres, features so high accuracy and is capable of working in real-time.Several classifiers were compared in the results section which gives additional knowledge about the system features.

Hardware Realization Of The MPEG-7 Edge Histogram Descriptor

The paper presents hardware implementation of the MPEG-7 edge histogram descriptor. The testing circuit was described using VHDL language and synthesized into FPGA. The RC1000 board with a Xilinx Virtex V1000 FPGA was chosen as the target platform. Experimental results of the descriptor efficiency are presented too

Embedded platform for local image descriptor based object detection

The article presents novel idea of a hardware accelerated image processing algorithm for embedded systems. The system is based on the well known Fast Retina Keypoint (FREAK) local image description algorithm. The solution utilizes Field Programmable Gate Array (FPGA) as a flexible module that is used to implement hardware acceleration of a given part of the image processing algorithm. The approach presented in this paper is slightly different. Since we are using very fast FREAK descriptor it is not our purpose to implement full feature extraction algorithm in hardware but just its most time-consuming part which is brute force matcher based on the Hamming distance. Moreover our goal was to design very flexible system so that the feature detection and extraction algorithm can be replaced without any interruption in the hardware accelerated part.

Multisensor data fusion using Elman neural networks

The paper presents a navigation system based on Elman Artificial Neural Network (ANN). The task of data fusion from different sensors is realized by trained ANN. Determining position in space is an issue of nonlinear hence. Not every type of ANN is used for such a task. Choice of Elman ANN was dictated by its construction and successfully applications to nonlinear problems requiring prediction. Elman network is composed of three layers. Comprises a layer of hidden layer units context which is connected to the hidden layer. Context-sensitive layer allows for store the values of previous hidden units. With this layer prediction is possible in sequential order. This is the effect of contextual memory where information is stored about what it was before. This kind of functionality is not able to provide any other standard neural network unidirectional. The system consists of MEMS (Micro Electro-Mechanical Systems) sensors, which are based on IMU (Inertial Measurement Unit). IMU is composed from gyroscopes, accelerometers and magnetometers which provide three dimensional linear accelerations and angular rates. This is a classic set of sensors for determining the position in space. The study presents the results of the implementation of algorithms for determining the position in space using trained Elman ANN. The data samples to train ANN were collected during the test flight of Quadrocopter. Paper presents the performance for different configurations of Elman ANN. Presented system provides easy addition of other sensors e.g. GPS/GLONASS receiver.

Real-Time Event Detection in Field Sport Videos

This chapter describes a real-time system for event detection in sports broadcasts. The approach presented is applicable to a wide range of field sports. Using two independent event detection approaches that work simultaneously, the system is capable of accurately detecting scores, near misses, and other exciting parts of a game that do not result in a score. The results obtained across a diverse dataset of different field sports are promising, demonstrating over 90 % accuracy for a feature-based event detector and 100 % accuracy for a scoreboard-based detector detecting only scores.

Trajectory realization with collision avoidance algorithm

This paper describes the approach to collision avoidance problem for 3-DOF anthropomorphic robot manipulators. The novelty of the approach is the decomposition of 3D space to two 2D spaces. Resulting is the computationally efficient algorithm, suitable for implementation in the real-time systems. Simulation of the anthropomorphic manipulator operating in three dimensional space with obstacles is also presented.

Investigation of generalization ability of a trained stochastic kinetic model of neuron

In this work we focus on the generalization ability of a biological neuron model. We consider a Hodgkin–Huxley type of biological neuron model, based on Markov kinetic schemes, trained with the gradient descent algorithm.