The advantage of an online semi-supervised boosting method which takes object tracking problem as a classification problem, is training a binary classifier from labeled and unlabeled examples. Appropriate object features are selected based on real time changes in the object. However, the online semi-supervised boosting method faces one key problem: The traditional self-training using the classification results to update the classifier itself, often leads to drifting or tracking failure, due to the accumulated error during each update of the tracker. To overcome the disadvantages of semi-supervised online boosting based on object tracking methods, the contribution of this paper is an improved online semi-supervised boosting method, in which the learning process is guided by positive (P) and negative (N) constraints, termed P-N constraints, which restrict the labeling of the unlabeled samples. First, we train the classification by an online semi-supervised boosting. Then, this classification is used to process the next frame. Finally, the classification is analyzed by the P-N constraints, which are used to verify if the labels of unlabeled data assigned by the classifier are in line with the assumptions made about positive and negative samples. The proposed algorithm can effectively improve the discriminative ability of the classifier and significantly alleviate the drifting problem in tracking applications. In the experiments, we demonstrate real-time tracking of our tracker on several challenging test sequences where our tracker outperforms other related on-line tracking methods and achieves promising tracking performance.
A high dynamic range infrared image of the sea surface scene includes the effects due to the sea clutter, mirror reflections from the wave facets, which decrease the visibility of the ship targets and their details. This paper provides an efficient adaptive enhancement technique for ship targets based on bilateral filtering and visual saliency detection. The 14- bit raw image is separated into a detail layer and a base layer by applying an adaptive bilateral filter. Then the two layers are processed separately and added afterward. Hereafter by employing visual saliency detection we can get the gain matrix to improve the contrast of ship targets. Finally, the image whose contrast is improved is quantized to the display range. The strength of our proposed method lies in its ability to inhibit the sea clutter and adaptability in different sea surface scene and shows a better performance in the contrast of the ship targets and the visibility of their details.
A target tracking model and a technique for target tracking filtering based on sequential unscented Kalman
filter are presented to improve target tracking performance of high resolution radar/infrared imaging sensor composite
guidance system. Firstly, a measurement model for imaging sensor based of the centroid of the target is derived from
images. Secondly, a measurement model for radar based of the centroid of the target is derived from traits of high
resolution radar. Finally, the data fusion filtering framework for target tracking based on sequential unscented Kalman
filter is presented. From the results of simulated experiments, average rate and target tracking accuracy of convergence
for the technique developed are superior to those of other techniques. In conclusion, the target tracking model and
filtering algorithm developed are proper for high resolution radar/infrared imagery sensor composite guidance system.
Most of edge detection algorithms are based on the gradient method. When texture is complex, especially in scale
light background, the image may be cluttered. Thus, it is hard to find useful information in the edge map. In this
article, a local energy model, which is relative to Human Vision System (HVS), is proposed to detect robust and
useful edges. We give non-classic receptive field (Non-CRF) methods to detect interested edges. The experimental
results show that the method we proposed in this paper is much better than gradient methods in detecting edges
In order to conquer the drawback of over-smoothness in the MRF model, a kind of discontinuity-adaptive Gaussian
Markov random field (DA-GMRF) model is defined, in which the edge information of image is used to construct
corresponding energy functions. After this a FLIR image segmentation method based on DA-GMRF model is proposed,
which includes initialization and optimization of the label field. A multi-threshold image segmentation algorithm based
on potential function of gray histogram is presented to initialize the label field. This algorithm can determine region
number and multi-thresholds automatically. Metroplis Sampler algorithm is adopted to optimize the label field.
Segmentation experiments on several images show that the algorithm proposed is effective.