Real Time Machine Learning Based Car Detection in Images With Fast Training

Our primary interest is to build fast and reliable object recognizers in images based on small training sets. This is important in cases where the training set needs to be built mostly manually, as in the case that we studied, the recognition of the Honda Accord 2004 from rear views. We describe a novel variant of the AdaBoost based learning algorithm, which builds a strong classifier by incremental addition of weak classifiers (WCs) that minimize the combined error of the already selected WCs. Each WC is trained only once, and examples do not change their weights. We describe a set of appropriate feature types for the considered recognition problem, including a redness measure and dominant edge orientations. The existing edge orientation bin division was improved by shifting so that all horizontal (vertical, respectively) edges belong to the same bin. We propose to pre-eliminate features whose best threshold value is near the trivial position at the minimum or maximum of all threshold values. This is a novel method that has reduced the training set WC quantity to less than 10% of its original number, greatly speeding up training time, and showing no negative impact on the quality of the final classifier. We also modified the AdaBoost based learning machine. Our experiments indicate that the set of features used by Viola and Jones and others for face recognition was inefficient for our problem, recognizing cars accurately and in real time with fast training. Our training method has resulted in finding a very accurate classifier containing merely 30 WCs after about 1 h of training. Compared to existing literature, we have overall achieved the design of a real time object detection machine with the least number of examples, the least number of WCs, the fastest training time, and with competitive detection and false positive rates.     

Adaptive Spatial Partitioning for Multidimensional Data Streams

We propose a space-efficient scheme for summarizing multidimensional data streams. Our sketch can be used to solve spatial versions of several classical data stream queries efficiently. For instance, we can track ε-hot spots, which are congruent boxes containing at least an ε fraction of the stream, and maintain hierarchical heavy hitters in d dimensions. Our sketch can also be viewed as a multidimensional generalization of the ε-approximate quantile summary. The space complexity of our scheme is O((1/ε) log R) if the points lie in the domain [0, R]^d, where d is assumed to be a constant. The scheme extends to the sliding window model with a log (ε n) factor increase in space, where n is the size of the sliding window. Our sketch can also be used to answer ε-approximate rectangular range queries over a stream of d-dimensional points.     

Hierarchical Evolution of High Definition Television

This paper presents an economically attractive system for the hierarchical evolution of High Definition Television (HDTV) compatible with the NTSC installed base. It describes the two main signals of this system: HDMAC-60 and HDNTSC. The first, HDMAC-60, is a feeder signal for use between the program origination and program redistribution points, and used also as a DBS emission format. The second, HDNTSC, is a delivery signal for high definition terrestrial broadcast, CATV distribution and VCR plus laser video disc applications. We propose this emission system as a means for transmitting HDTV pictures to consumers on the North American continent.     

Physical nature of empirical time dependencies of the strength and failure characteristics of materials

Correlations between each of two parameters of empirical relationships describing arrays of experimental data on the long-term strength, creep, multi- and low-cycle fatigue, and the growth of fatigue cracks are established. Well-defined relationships between parameters within each array make it possible to reduce significantly the required volume of material tests. The physical essence of the observed correlations is associated with the thermally active nature of damage processes, which accompany each form of material tests. It is demonstrated that the pattern of the dependence of correlating empirical parameters on fundamental characteristics of thermally activated damage processes can be established for all types of tests under consideration: the height of the potential barrier, the frequency factor, etc. For all types of tests, analytical predictions are in good qualitative agreement with correlations observed in experiment.Translated from Problemy Prochnosti, No. 6, pp. 3–9, June, 1994.     

On the role of hydrophilicity and hydrophobicity in aqueous heterophase polymerization

This paper reports on experimental results of aqueous heterophase polymerizations with monomers of quite different solubility in water ranging from the water-soluble 2-hydroxyethyl methacrylate to lauryl methacrylate with solubility in water of only about 10⁻⁴ mM. A calorimetric study revealed the strong influence of both the hydrophilicity of the monomer and the stirrer speed on the rate of polymerization in the absence of surfactants. In order to obtain maximum latex yield and high efficiency (which is a measure considering colloidal properties, polymerization recipe, and polymerization parameters) the initiator-surfactant combination must be properly chosen in dependence on the hydrophilicity of the monomer. Results are presented for sodium alkyl sulfates or disodium-N-stearoyl-l-glutamate as surfactant and potassium peroxodisulfate, or poly(ethylene glycol)-azo- compounds, or 2,2′-azobis(N-2′-methylpropanoyl-2-amino-alkyl-1)-sulfonates as initiators.     

An alignment-free fingerprint bio-cryptosystem based on modified Voronoi neighbor structures

Bio-cryptography is an emerging security technology which combines cryptography with biometrics. A good bio-cryptosystem is required to protect the privacy of the relevant biometric data as well as achieving high recognition accuracy. Fingerprints have been widely used in bio-cryptosystem design. However, fingerprint uncertainty caused by distortion and rotation during the image capturing process makes it difficult to achieve a high recognition rate in most bio-cryptographic systems. Moreover, most existing bio-cryptosystems rely on the accurate detection of singular points for fingerprint image pre-alignment, which is very hard to achieve, and the image rotation transformation during the alignment process can cause significant singular point deviation and minutiae changes. In this paper, by taking full advantage of local Voronoi neighbor structures (VNSs), e.g. local structural stability and distortion insensitivity, we propose an alignment-free bio-cryptosystem based on fixed-length bit-string representations extracted from modified VNSs, which are rotation- and translation-invariant and distortion robust. The proposed alignment-free bio-cryptosystem is able to provide strong security while achieving good recognition performance. Experimental results in comparison with most existing alignment-free bio-cryptosystems using the publicly-available databases show the validity of the proposed scheme.     

Automatic player position detection in basketball games

This paper presents us with a framework for the automatic player position detection (APPD) in the game of basketball. Court players are detected in the images broadcasted via television stations. In them, at any point of time, the view is from only one camera. This makes the detection process much more difficult. The player detection is based on the mixture of non-oriented pictorial structures. The detection of body parts is performed by the Support Vector Machine (SVM) algorithm. The results of these detections are combined together with constraints on their locations, which specify the position of one body part with respect to the parent body part. In order to train the whole model, we used a latent form of SVM called the latent SVM (LSVM). Such approach generated the statistical accuracy of about 82 %, which represents one of the best results in basketball player detection framework. Beside players, the algorithm detected a certain number of false positive objects. These are mostly people from the audience and the referees as well. This paper contains a simple and robust solution to remove them all, based on the play court boundaries and the histogram comparison. Separating players in different teams is done by k-means clustering. The inputs to this algorithm are saturation histograms calculated on the jerseys. A spatial transformation is determined by the detected play court boundaries and the actual court measures. Using this transformation, points representing the location of detected players in TV images are mapped to the actual location of players on the court, which was the main goal of our research. The proposed solution is sound and efficient. In addition, it is backed up by the experimental results obtained using the model of the actual footage of basketball games.