Search

Non-rigid object tracking is an important yet challenging task in computer vision. In this paper, a multi-patch neural network (MPNet) model is presented to address the problem of non-rigid object tracking. The model learns a multiple patch based framework, which mainly consists of two branches of neural networks. One branch is to track the global target and the other branch partitions the target into multiple patches which are tracked separately. The global tracking and the multiple patch tracking are combined to compute the final tracking results. Compared with the existing methods, our model exploits the trajectories of various parts of a non-rigid object and therefore can accurately track the non-rigid object. Experiments on visual tracking datasets prove the strength of the proposed method.

Anomaly detection on system logs is to report system failures with utilization of console logs collected from devices, which ensures the reliability of systems. Most previous researches split logs into sequential time windows and regarded each window as an independent instance for classification using popular machine learning methods like support vector machine(SVM), however, neglected the time patterns under logs. Those approaches also suffer from information loss due to the vector representation, and high dimensionality if there is a large number of log events. To make up these deficiencies, unlike most traditional methods that used a vector to represent a period behavior at the macro level, we construct a 2D matrix to reveal more detailed system behaviors in the time period by dividing each window into sequential subwindows. To provide a more efficient representation, we further use the ant colony optimization algorithm to find a highly-coupled event template as the horizontal index of the 2D window matrix to replace the disordered one. To capture time dependencies, a multi-module convolutional auto-encoder is configured as that different paralleled modules scan among different time intervals to extract information respectively. These features are then concatenated in latent space as the final input, which contains diversified time information, for classification by SVM. The experiments on Blue Gene/L log dataset showed that our proposed method outperforms the state-of-art SVM method.

The phase function of an SAR signal in the two dimensional wavenumber domain is usually derived by using the principal of stationary phase. From this phase function, the phase error of the conventional range-Doppler algorithm when used for high resolution large squint angle SAR imaging is obvious [Bamler 1992]. The secondary range compression and other approximate wavenumber domain algorithm can be derived based on the phase function. But the physical meaning of the secondary range compression can not be directly perceived. In this paper we derive the phase function from SAR imaging geometry by using the Doppler-azimuth coordinate relationship in different RF frequency. From this point of view, it is easy to understand the phase error originated from the range-Doppler algorithm.

The radar cross section (RCS) of sea surface measured by a radar altimeter is dominated by specular reflection. There are two factors affecting the normalized RCS of this type of target: one is the mean surface slope; the other is the radar look angle. For nadir looking altimeter, a long wave length wave has the effect of changing the local incident angle of the coming electromagnetic waves. The authors investigate the possibility of using synthetic aperture radar altimeter for estimation of directional wave spectrum. Synthetic aperture processing inherently can give a high spatial resolution in the along track direction by combining of several pulses. Then by appropriate sampling within each pulse they can sample the sea surface at high resolution in the across-track direction. They study this novel process by using computer simulation. They simulate the image got by synthetic aperture radar altimeter. The resolution achieved by SAR altimeter in the simulation is about 25 m/spl times/25 m. The results show they can estimate the directional wave spectrum but with an ambiguity in wave direction.