V. S. Kirichuk

Real-time detection of moving objects from a sequence of images

A new algorithm of estimating the stationary part of the background of the observed scene and its adaptation to changes in this scene and observation conditions are proposed. The algorithm is optimal in the statistical sense and employs the minimum number of statistical characteristics (moving average, variance, and median), which allows their recurrent calculation necessary for real-time implementation. The validity of the proposed method is confirmed by results of a computational experiment with real video sequences.

Construction of a multichannel filter for detecting point targets in the image formed by a matrix photodetector

A new approach is proposed to construct a multichannel filter for detecting small-size targets in images recorded by a matrix photodetector. For estimating the pulse response of each channel of the filter, the set of experimentally obtained images of targets having different locations with respect to photodetector cells is divided into several subsets, which are used as reference elements in calculating the pulse response of the filter. Experimental data are presented, which demonstrate appreciable reduction of the probability of missing the targets in images subjected to processing with a filter constructed by the proposed method.

Space-time analysis of video sequences for detecting abandoned objects

A new algorithm for detecting abandoned objects is proposed. The algorithm is based on analyzing the space and time characteristics of the overall background, background consisting of stationary objects, and current image of the video sequence. The algorithm can operate under conditions of short-time overlapping of the abandoned object, has a moderate computational complexity, and is designed for application in video surveillance systems.

Model of a persistent photodetector for analyzing the threshold sensitivity in the problem of detection of a moving point object

A model of the direction-finding characteristic of a moving point object for a persistent photodetector is proposed. The model ensures the calculation of the direction-finding characteristic as a function of photodetector parameters and object velocity, which form the basis for calculating the probability of detection of moving point objects. Results of numerical experiments are reported.

Specific features of operation of fallen person detection algorithms based on a sequence of scene images

Specific features of operation of fallen person detection algorithms having a moderate computational complexity and based on a sequence of images obtained by stationary video surveillance cameras are analyzed. Approaches to enhancing the algorithm operation stability are proposed. One variant of obtaining a more reliable silhouette of the object is considered; parameters used in recognition of the object states are analyzed and selected. A method of increasing the detector stability is proposed. This method is based on identifying intervals of stability of the average values of parameters, which allows reliability of detection of the object states to be improved. Results of computational experiments are presented.

Fast algorithm of detection of boundary points in images

A method of detecting boundary points in brightness images, based on subpixel calculation of the brightness difference, is proposed. In a fragment 4×4 pixels in size, this method allows calculating 12 directions of the jump in brightness; the algorithm complexity is ∼34N atomic operations. The method considered is compared with available methods of detection of boundary points in the image. The algorithm proposed is demonstrated to be more stable to the “salt and pepper” noise, ensures more stable determination of the brightness jump direction, and provides a more intense response to the signal. An application of the method is noted.

Modeling and analysis of motion of biological objects on the basis of a sequence of images obtained in studying the motion activity

A new approach to analyzing the data obtained by tracking biological objects in studying the motion activity is considered. The approach is based on a proposed mathematical model of motion of a biological object. The object behavior observed during the experiment can be described in the form of model parameters. A numerical experiment with model data is performed, which shows that the estimates predicted by the proposed method ensure a smaller root-mean-square deviation of the object state classification than other available methods.

Estimate of the effect of antidepressants in the forced swimming test

Justification and experimental verification of the computer vision algorithm and the maximum likelihood method for automatic estimation of the time of animal immobility in the forced swimming test are given. The computer vision algorithm is based on studying the difference between the frames with the animal images. The immobility estimates obtained by the existing methods and on the basis of the proposed criterion are compared.

Algorithm of image reconstruction in the problem of object detection during circular microscanning

A model of image recording in the circular microscanning regime is considered. An algorithm that allows obtaining an image with a large number of samples from a set of smaller-size frames for a known path of the scanning array is proposed. Numerical simulations are performed to estimate the changes in the amplitude of a small-size object with a static position of the photodetector array and microscanning in the start/stop mode. The influence of the gap between the array elements and scanning radius on the object amplitude is studied. The final ratio of the object amplitude to the random noise level in the case of static accumulation and circular microscanning in the start/stop mode are compared for arrays with different sizes of the gap between the array elements.

Suppression of a quasi-stationary background in a sequence of images by means of interframe processing

Various methods of background compensation in a sequence of images are analyzed under conditions where the main change in the background is its spatial shift. Methods of estimating the background in a current frame by means of interpolation of data contained in the previous frames are considered. Analytical estimates of the compensation error are given as functions of the spectral properties of images and interpolation methods. Results of numerical experiments are reported, which confirm these analytical estimates and show that the use of local interpolating filters allows extremely low errors of compensation.