Csaba Rekeczky

Bi-i: a standalone ultra high speed cellular vision system

The Bi-i standalone cellular vision system is introduced and discussed. In the first part, the underlying sensor and system level architectures are presented and various implementations are overviewed. This computing platform consists of state-of-the-art sensing, cellular sensing-processing, digital signal processing and communication devices that make it possible to use the system as an ideal computing platform for combined topographic and non-topographic calculations in sensing-processing-actuation scenarios. In the second part of the paper, ultra-high frame rate laboratory experiments are shown and discussed to highlight the most peculiar features of the system and its applicability in various industrial quality control areas. The overview underlines the potentials of the Bi-i vision system for unmanned intelligent vehicle applications in visual exploration, identification, tracking and navigation.

Topographic cellular active contour techniques: theory, implementations and comparisons

This paper overviews some massively parallel topographic cellular computational approaches proposed for contour localization and tracking. When implemented on a focal plane cellular array microprocessor, these algorithms offer real-time object contour localization and tracking—even at very high frame rates. Three specific methods (Constrained Wave Computing, Pixel Level Snakes and Moving Patch Method) will be described and compared along with their associated hardware–software architectures. Computational complexity, implementation, and performance related issues are discussed on a common platform (ACE-BOX with the ACEx CNN-UM chips). In conclusion, a novel architecture is proposed incorporating the best solutions learned from this comparative study. Copyright

Various implementations of topographic, sensory, cellular wave computers

The cellular wave computer architecture, based on the CNN universal machine principle, has been implemented recently in many different physical forms. The mixed mode CMOS, the emulated digital (cell wise or as aggregated arrays), FPGA, DSP, as well as optical implementations are the main examples. In many cases, the sensory array is integrated as well. The new self contained unit, called Bi-i, winning the product of the year title at the Vision 2003 in Stuttgart as the fastest camera-computer, shows the application interest and impact being capable of sensing-computing with 50000 frames per second. In this paper a clear and concise comparison is presented between the various implementation modes.

Elastic Grid Based Analysis of Motion Field for Object-Motion Detection in Airborne Video Flows

The visual navigation system of a UAV is a complex embedded device designed to modify the path of the platform depending on objects or events detected on the ground. In the visual field of the autopilot these events could be formalized as specific space-time signatures. Processing all pixels captured by the on-board camera(s) in real time with high frame rate needs huge computational effort that is often unnecessary. An adequate computational strategy would focus on the interesting locations only as in the visual system of various species. In this article we describe an automatic focusing mechanism relying on optical flow calculation for detecting moving objects on the ground, thus efficiently separating the motion of interest from ego-motion of the platform.

Elastic Grid-Based Multi-Fovea Algorithm for Real-Time Object-Motion Detection in Airborne Surveillance

In this chapter, a generic multi-fovea video processing architecture is presented, which supports a broad class of algorithms designed for real-time motion detection in moving platform surveillance. The various processing stages of these algorithms can be decomposed into three classes: computationally expensive calculations can be focused onto multiple foveal regions that are selected by a preprocessing step running on a highly parallel topological array and leaving only the nontopological (typically vector-matrix) computations to be executed on serial processing elements. The multi-fovea framework used in this chapter is a generalized hardware architecture enabling an efficient partitioning and mapping of different algorithms with enough flexibility to achieve good compromise in the design tradeoff between computational complexity versus output quality. We introduce and compare several variants of four different classes of state-of-the-art algorithms in the field of independent motion analysis and detection. On the basis of the analysis, we propose a new algorithm called the Elastic Grid Multi-Fovea Detector characterized by moderate hardware complexity while maintaining competitive detection quality.

Low-Power Processor Array Design Strategy for Solving Computationally Intensive 2D Topographic Problems

2D wave type topographic operators are distributed into six classes, based on their implementation methods on different low-power many-core architectures. The following architectures are considered: (1) pipe-line architecture, (2) coarse-grain cellular parallel architecture, (3) fine-grain fully parallel cellular architecture with discrete time processing, (4) fine-grain fully parallel cellular architecture with continuous time processing, and (5) DSP-memory architecture as a reference. Efficient implementation methods of the classes are shown on each architecture. The processor utilization efficiencies, as well as the execution times, and the major constrains are calculated. On the basis of the calculated parameters, an optimal architecture can be selected for a given algorithm.

Online 3-D Reconstruction of the Right Atrium From Echocardiography Data via a Topographic Cellular Contour Extraction Algorithm

A computational method providing online, automated 3-D reconstruction of the right atrium of the human heart is presented in this paper. Endocardial boundaries were extracted from transesophageal ultrasound data by a topographic cellular contour extraction (TCCE) algorithm. The TCCE method was implemented on a continuous-time, analog, massively parallel processor, and on a digital serial processor. Processing speeds were 500 or 40 frames per second, depending on the hardware used. Extracted boundary point sets were rendered into a 3-D mesh and the volume of the cavity was quantified. Accuracy of volume quantification was validated on 60 in vitro static phantoms and 12 clinical subjects. For the clinical recordings, reference volumes were estimated from manually delineated endocardial boundaries. The average error of volume quantification by the TCCE method was 8% ±5% (n = 12), the average of the interobserver variability between two independent human experts was 5% ±2% (n = 6). Interactive planning of atrial septal defect closure in pediatric cardiology is presented as an example that demonstrates a potential clinical application of the method.

Sensor Integration in Autonomous Systems

In this survey introduction to the special session of the same title at the IEEE International Symposium on Circuits and Systems, we review past work in the development of bio-inspired circuits and platforms for sensing and actuation, as well as recent work that has been done in integrating these systems into applications.

CELLULAR MULTI-ADAPTIVE ANALOGIC COMPUTING: ALGORITHMS FOR UAV APPLICATIONS

An efficient adaptive algorithm in process real-time applications should make optimal use of the available computing power for reaching specific design goals. Relying on appropriate strategies, spatial resolution/temporal rate can be traded against computational complexity; and sensitivity traded against robustness in an adaptive process. In this work, we present an algorithmic framework where spatial multi-grid computing is placed within a temporal multi-rate structure, and at each spatial grid point, the computation is based on an adaptive multi-scale approach. The algorithms utilize an analogic (analog and logical) architecture consisting of a high-resolution optical sensor, a low-resolution cellular sensorprocessor (cellular nonlinear network – CNN - based chip) and a digital signal processor. The proposed framework makes the acquisition of a spatially and temporally consistent image flow possible even in case of extreme variations in the environment. It ideally supports the handling of difficult problems on a moving platform such as terrain identification, navigation parameter estimation and multi-target tracking. The proposed spatio-temporal adaptation relies on a feature based optical flow estimation that can be efficiently calculated on available CNN chips. The quality of the adaptation is evaluated compared to non-adaptive spatio-temporal behavior (when the input flow is over-sampled and therefore results in redundant data processing with the unnecessary wasting of computing power). We demonstrate how multi-channel visual flow analysis and a classifier driven visual attention-selection mechanism can be efficiently supported by an analogic architecture. We also use a visual navigation example for recovering the yaw-pitch-roll parameters from motion field estimates in order to analyze the proposed adaptive algorithmic framework. The experiments performed on an analogic CNN hardware prototype will highlight the application potentials for unmanned air vehicle (UAV) applications.

Cellular Multi-core Processor Carrier Chip for Nanoantenna Integration and Experiments

A new generation of IR and sub-millimeter wave radiation detector imager array with integrated per channel high-gain capacitive amplifiers and digital signal processing/enhancement circuitry was designed. The multi-core processor carrier chip, with the analog interface and the digital processor array were implemented in standard 0.18-μm CMOS technology and verified within a compact measurement system. Characterization with external photosensors has been completed and the associated measurement results are presented. A concept for nano antenna type detector array integration to the processor carrier was also developed, and some preliminary experiments have been conducted with metal-oxide-metal (MOM) diodes.