Synthetic aperture radar (SAR) is a powerful tool for remote sensing of the Earth surface. In the paper, several applications of pattern detection and recognition algorithms for extraction of information from SAR images are discussed. In particular, an idea of usage of optical flow techniques for automatic estimation of the moving target displacements from a sequence of single-look SAR images is proposed. It is shown, how this technique can be adopted for SAR imagery. In addition, it is demonstrated that local feature descriptors can be used for automatic SAR panorama creation without information about reference platform orientation. Finally, a problem of automatic target recognition is analyzed. It is shown, that a fusion of azimuth and range profiles of the ground target allows to achieve a competitive accuracy using quite compact feature vectors.
Category: Research
Focusing of Static and Moving Targets in Real Flight Conditions using SAR Technique
Synthetic aperture radar is a very popular and widely used instrument for various remote sensing tasks. One of the most challenging problems is to obtain high-quality images in the case of unstable flight conditions. In the paper the problem of full platform motion compensation is discussed. A particular attention is given to the analysis of moving targets. Algorithm for estimation of moving target parameters is developed. Experimental results with real data are illustrated.
Autofocusing SAR Images via Local Estimates of Flight Trajectory
High-resolution imaging with an airborne synthetic aperture radar (SAR) calls for precise trajectory measurements that can hardly be achieved with common navigation systems. In this paper, an efficient method called the local-quadratic map-drift autofocus is developed for the estimation of residual (uncompensated) motion errors directly from the received radar data. The map-drift autofocus is applied locally on short time intervals to estimate the cross-track components of the aircraft acceleration. The estimated acceleration is then integrated to evaluate the residual trajectory errors on the whole data frame interval. The method has been successfully tested with an X-band airborne SAR system.
A Novel Algorithm for Estimation of Moving Target Parameters with Single-Antenna SAR
The idea for automatic extraction of the moving target parameters using single-antenna synthetic aperture radar is proposed. The displacements of the target are estimated from the sequence of the single-look SAR images using the Lucas-Kanade optical flow technique. It is shown that both moving target location in the SAR look and its shift
direction can be extracted. In addition, the linear road segments are automatically detected from the multi-look SAR image using the Canny edge map and local pixels intensity gradient analysis. Proposed ideas are incorporated into the novel algorithm capable to unambiguously extract the moving target parameters using the single-antenna SAR system. Experimental examples obtained with X-band airborne SAR system are illustrated as well.
Frame-Based SAR Processing and Automatic Moving Targets Parameters Extraction
Synthetic aperture radar is a very popular and widely used instrument for various remote sensing tasks. In the paper, we propose several novel ideas for improvement of the efficiency of the modern SAR systems. At first, the problem of the automatic image stitching is considered. Instead of the common cross-correlation based solution the local features detection and description techniques are proposed. Secondly, we analyze the problem of the moving target parameters estimation. It is shown that the optical flow techniques can be used for the automatic extraction of the moving target shifts from the sequence of SAR looks. Experimental examples with real SAR data are illustrated and comprehensively discussed.
Efficient Two-Step Approach for Automatic Number Plate Detection
Intelligent transportation systems are rapidly growing mainly due to active development of novel hardware and software solutions. In the paper a problem of automatical number plate detection is considered. An efficient two-step approach based on plate candidates extraction with further classification by neural network is proposed. Stroke width transform and contours detection techniques are utilized for the image preprocessing and extraction of regions of interest. Different local feature sets are used for the final number plate detection step. Efficiency of the developed method is tested with real datasets.
Robust Number Plate Detector Based on Stroke Width Transform and Neural Network
The number plate detection is a key step affecting the overall performance of the number plate recognition system. In the paper a novel algorithm for this purpose is proposed. The approach is based on the detection of text areas using the stroke width transform. More plate candidates are detected using the specifically developed image preprocessing scheme based on set of morphological operators and contour analysis. The final number plate candidates are properly classified using the neural network which is learned from the training dataset. Experiment results indicate on the high performance of the developed methodology
Efficient Data Focusing and Trajectory Reconstruction in Airborne SAR Systems
Formation of high-resolution SAR images from light-weight platforms is a challenging task primarily due to high instability of such platforms. Additional difficulties are related with the precision of navigation systems. In the paper the problem of residual trajectory deviations are analyzed. An efficient trajectory reconstruction method is proposed. Important practical aspects of the developed approach are discussed. Considered ideas are incorporated into the SAR processing chain. The algorithm efficiency is proven via examples of real SAR data obtained with an X-band airborne SAR system.
