Detection and tracking of rotated and scaled targets by use of Hilbert-wavelet transform

Image rotation measurement in scene matching based on holographic optical correlator

Based on the stationary random properties of remote sensing images, a correlation model is proposed to resolve the effects of the image rotation and translation on the correlation value in scene matching. The rotation invariance is achieved by measuring the image rotation with the model and compensating the rotation before the 2D translation scene matching. The input image is rotated from -5° to 5° at an interval of 1° and 11 new images are generated. The 11 new images correlate with all the template images and eleven correlation matrices are obtained. The maximum values of each correlation matrix are picked up and they could follow a fixed curve predicted by the model. Fitting the curve, the rotation corresponding to the estimated peak of the curve is considered to be the rotation of the input image. The rotation measurement of the input image can be as accurate as 0.05°. With an extra 36 rotations of the input image, the measuring range of rotation can be enlarged into ±180°. This method could be very fast and accurate for scene matching in the parallel multichannel holographic optical correlator.

Detection and tracking of sea-surface targets in infrared and visual band videos using the bag-of-features technique with scale-invariant feature transform

Sea-surface targets are automatically detected and tracked using the bag-of-features (BOF) technique with the scale-invariant feature transform (SIFT) in infrared (IR) and visual (VIS) band videos. Features corresponding to the sea-surface targets and background are first clustered using a training set offline, and these features are then used for online target detection using the BOF technique. The features corresponding to the targets are matched to those in the subsequent frame for target tracking purposes with a set of heuristic rules. Tracking performance is compared with an optical-flow-based method with respect to the ground truth target positions for different real IR and VIS band videos and synthetic IR videos. Scenarios are composed of videos recorded/generated at different times of day, containing single and multiple targets located at different ranges and orientations. The experimental results show that sea-surface targets can be detected and tracked with plausible accuracies by using the BOF technique with the SIFT in both IR and VIS band videos.

Infrared target tracking with kernel-based performance metric and eigenvalue-based similarity measure

An infrared target tracking framework is presented that consists of three main parts: mean shift tracking, its tracking performance evaluation, and position correction. The mean shift tracking algorithm, which is a widely used kernel-based method, has been developed for the initial tracking for its efficiency and effectiveness. A performance evaluation module is applied for the online evaluation of its tracking performance with a kernel- based metric to unify the tracking and performance metric within a kernel-based tracking framework. Then the tracking performance evaluation result is input into a controller in which a decision is made whether to trigger a position correction process. The position correction module employs a matching method with a new eigenvalue-based similarity measure computed from a local complexity degree weighted covariance matrix. Experimental results on real-life infrared image sequences are presented to demonstrate the efficacy of the proposed method.