Indirect time-of-flight measurement with SPADs is performed by counting incident photons in several time windows. Since SPADs exhibit dead time not all incident photons can be counted within a given time window. This affects the expected values and, hence, the variance of the distance measurement. For photon detection rates close to the inverse of the dead time, which defines the maximum count rate of a SPAD, the probability of photon detection cannot be assumed constant within the window anymore. In this paper the effects of dead time on the photon counts are analyzed by employing statistical calculations. Based on these a model to correct such effects can be derived.

Systems which perform digital signal processing in continuous-time are attractive for a number of applications like biomedical implants, hearing aids, remote sensors, telecommunications, and audio and speech processing. A main difference to sampled data systems is the realization of the delay elements which must be implemented as quasi-continuous time delay lines. Thus a large chip area is required for the delay elements which are also a critical point of these systems. Therefore in this paper a method will be presented which allows reducing the implementation costs and power consumption of these elements. This can be achieved by granularity reduction without sacrificing performance. Furthermore implementation aspects for an integrated solution will be covered.

We present an integrable solution for the detection of defective sensor elements using sigma-delta-(/spl Sigma//spl Delta/)-modulation and a matched filter. The sensor element is stimulated using a pseudo random binary sequence (PRBS). The sensor signal is read out and the analog output is digitized using a /spl Sigma//spl Delta/-modulator. The binary pulse density stream of the /spl Sigma//spl Delta/-modulator is the output of the sensor system and thus should ideally contain the PRBS. A matched filter has the task of detecting the pseudo random sequence in the pulse density stream and its sampled output is compared to a threshold thus making it possible to judge the functionality of the sensor element. By evaluating the magnitude of the matched filter output it is also possible to measure the sensor sensitivity. We present a discrete solution of this method, but an integrated chip using a standard 1.2 /spl mu/m CMOS process has been designed and is being fabricated.

In this paper we present a technique for nonlinear function generation which uses a novel sparse look-up table approach. This novel approach can be derived from the simple look-up table concept and utilizes the interpolating properties of oversampled /spl Sigma//spl Delta/-modulators. We show that the approximation process with oversampled /spl Sigma//spl Delta/-modulators can be generally understood as a weighted sum of piecewise-continuous polynomial B-splines. The weights are digitally stored in a sparse look-up table memory which can be used for the nonlinear function mapping. The shape of the polynomial B-splines depends on the order of the loop-filter which is used in the /spl Sigma//spl Delta/-modulator topology. Derailed investigations of the approximation behaviour show that a spectrum of linear, square, and cubic piecewise-continuous interpolation can be achieved if we use up to 3rd order /spl Sigma//spl Delta/-modulators. With the increased order of modulator topology the approximation quality is also improved under the assumption that we use the same weight set for function mapping. The fact that polynomial B-splines delivers only nonzero values within a limited range of input arguments makes this approximation method much less sensitive to local errors than classical approximation methods like Lagrange or Newton interpolation.

In this paper we address the problem of detection and tracking of moving objects for surveillance or occupant detection systems. The primary goal in this framework is the motion estimation of the extracted foreground. To overcome the drawbacks characteristic of classical block matching techniques, this contribution presents a new feature based hierarchical locally adaptive multigrid (HLAM) block matching motion estimation technique based on a foreground detection procedure using an adaptive recursive temporal lowpass filter. It leads to a robust and precise motion field estimation, close to the true motion in the scene. The simulation results highlight the superior performance of the proposed method. It yields better performance than the classical exhaustive search and the modified three-step search (MTSS) technique in terms of the peak signal-to-noise ratio (PSNR).

In this communication we present a low cost smart airbag solution based on a high speed active CMOS camera system. The system continuously monitors the seats and quickly determines the occupancy status, and a passenger's position and size before the airbag is deployed. The camera uses an image sensor fabricated in a 1 /spl mu/m standard CMOS technology and is capable to acquire more than 1000 frames per second. Disturbing temporal instationarities in the captured images caused by illumination variations resulting from lighting condition changes while capturing the scene, and intrinsic motions of the objects in the scene are suppressed effectively by differencing images captured alternately with and without active structured illumination.

This paper presents an unsupervised texture segmentation algorithm based on feature extraction using multichannel Gabor filtering. It is shown that feature contrast, a criterion derived for Gabor filter parameter selection, is well suited for feature coordinate weighting in order to reduce the feature space dimension. The central idea of the proposed segmentation algorithm is to decompose the actual segmented image into disjunct areas called scrap images and use them after lowpass filtering as additional features for repeated k-means clustering and minimum distance classification. This yields a classification of texture regions with an improved degree of homogeneity while preserving precise texture boundaries.

This article describes a novel approach to orientation and scale-invariant detection of textured objects in images. It performs both, a segmentation of multi-object scenes and the identification of rotation angles and scale rates of textures in an image by applying a comparison with reference texture features stored in a database. The main novelty of the proposed method is the transform of rotation and dilation into shifts in the feature space by employing a polar-log Gabor filter bank. Texture segmentation and identification of the rotation angles and scale rates have been carried out using symmetric phase only matched filters. The simulation results illustrated highlight the performance of the presented method in an exemplary manner.

In this paper we present a new biology-inspired neuron model and its real-time realization using a dedicated neural hardware emulator. The biological neuron model overcomes the limitations of classical neuron models by including dynamic features such as adaptive synaptic delays. The emulator used for its realization is based on a special communication processor optimized for the global exchange of pulse messages between neuron processors. The use of the model together with the emulator in real-time adaptive signal processing is shown using an example in the field of fault-tolerant adaptive beamforming.

Recent studies have confirmed that the multichannel Gabor decomposition represents an excellent tool for image segmentation and boundary detection. Unfortunately, this approach when used for unsupervised image analysis tasks imposes excessive storage requirements due to the nonorthogonality of the basis functions and is computationally highly demanding. In this correspondence, we propose a novel method for efficient image analysis that uses tuned matched Gabor filters. The algorithmic determination of the parameters of the Gabor filters is based on the analysis of spectral feature contrasts obtained from iterative computation of pyramidal Gabor transforms with progressive dyadic decrease of elementary cell sizes. The method requires no a priori knowledge of the analyzed image so that the analysis is unsupervised. Computer simulations applied to different classes of textures illustrate the matching property of the tuned Gabor filters derived using our determination algorithm. Also, their capability to extract significant image information and thus enable an easy and efficient low-level image analysis will be demonstrated.

This work describes a parallel neural network emulator which uses standard DSPs and application-specific VLSI communication processors with an integrated hardware routing algorithm. The use of DSPs as programmable processing elements enables the emulation of different types of neurons including biologically inspired models with learnable synaptic weights and delays, variable neuron gain, and static and dynamic thresholding. Locally interconnected communication processors attached to each DSP can span up a 2D- or 3D-computing grid and thus form a highly parallel network topology capable of global packet switched routing.>

We present an unified method that can be used to compute analytically the coefficients of the discrete non-orthogonal Gabor expansion for arbitrary type of orthogonal kernel. Based on the proposed algorithm an architecture is derived that enables a fast computation of modified time-frequency transforms such as Gabor-DCT. It is shown that for the special case of equidistant elementary cells the generalized Gabor transform can be modeled as a conventional block transform of modulated and filtered input signals. From this point of view it follows that the biorthogonal analysis window which is used in the modulation necessary for computation of the transform coefficients is independent of the selected orthogonal kernel. The problems caused by the non-stationarity of the modulated and filtered signal are discussed and the coding performance and energy compaction of the classical Gabor transform, the Gabor-DCT, and the well known block DCT are compared.>

New algorithms called interlaced-correlated-double and triple-sampling are presented that perform noise reduction in analog discrete-time circuits. Both are based on multiple correlated-double-sampling principle. Theoretical improvement of noise reduction has been verified using computer simulations. The contribution also discusses CMOS realization of the proposed method.>

One of the main problems of the CORDIC algorithm is the limited convergence domain, in which the functions can be calculated. Two different approaches can be employed to overcome this constraint: first, an argument reduction method and, second, an expansion of the CORDIC convergence domain. While the first approach requires significant processing overhead due to the need for divisions especially for tanh/sup -1/, the second technique achieves an increased but still limited convergence domain only. In this brief contribution, we present a unified division-free argument reduction method and a regular pipeline/array architecture for floating point or fixed point implementations which results in savings of computation time. In contrast to previous methods we avoid extra CORDIC arithmetic for realization of argument reduction.>

An algorithm that computes the Gabor coefficients of an image is presented. An adaptive filter that uses the complex least mean-square algorithm for their computation is proposed, and its numerical characteristics are discussed. It is shown that the filter is stable under certain conditions. Because the Gabor transformation seems to be an excellent tool for image compression, the efficiency of information coding using the Gabor coefficients is investigated and compared with coding that uses the coefficients of the discrete cosine transformation (DCT). Properties of both transformations are discussed.

Several methods for increasing the speed of the CORDIC algorithm are presented. First, an improved method which guarantees a constant scale factor when employing redundant addition schemes is developed. Then, an architecture with increased parallelism which considerably reduces the CORDIC latency time and the amount of hardware is described.>

In this paper a statistical approach for pattern recognition, based on a distance transformation, will be presented. The algorism enables multiclass rotation- and translation invariant pattern recognition and can be made optionally scale invariant. Beside the theory of the algorithm we will present some practical application examples and propose a system architecture for a high speed hardware implementation including an ASIC architecture.

It is the purpose of this paper to compare analog and digital semiconductor circuits. Using simple models, performance of these circuits will be evaluated as far as signal-to-noise ratio, power dissipation, bandwidth, and system capacity are concerned.