A Moving Object Tracking Scheme Based On Compressive Correlated Imaging

Correlated imaging, also named ‘ghost imaging’, is widely applied to medical images, military affairs, distributed image processing and communications for its unique non-locality and high resolution. With the development of correlated imaging scheme, the practical application about how to track moving objects using compressive correlated imaging draws more and more attentions. The thesis studies the way to improve the speed of compressive correlated imaging, as well as the tradeoff between imaging speed and imaging quality.First of all, the paper focuses on the speed of imaging in tracking moving objects based on compressive correlated imaging. To achieve this goal, the measurement matrix of large scale is adopted in the compressive correlated imaging for the correlation between the adjacent pixels and a mean shift tracking algorithm is used to track the moving objects. The simulation shows that:(1)By using the measurement matrix of appropriate scale, a clearer outline of images can be achieved with fewer number of measurements and less reconstruction time.(2)For tracking the moving objects, the effect of mean shift tracking algorithm is not good when the object is similar to the background. For instance, for “lena” image with the size of 128 * 128, when the scale of measurement matrix takes 4, the image can be reconstructed with the compression ratio of 0.06, while for “car” image with the size of 64* 128, only a compression ratio of 0.04 is needed to recover the image under the same conditions. Furthermore, using the mean shift algorithm, the tracking performance of the latter is better than the former.Secondly, the problem of the imaging quality in tracking moving objects based on compressive correlated imaging is considered. Since the difference between the measurement results in compressive correlated imaging could be transferred to the difference operation on the original objects, three frame difference compressive correlated imaging scheme is proposed to balance the same part between successive frames. In addition, the image which filters out the background could be tracked with the methods of projection positioning and locus fitting. The simulation shows that:(1)By using three frame difference compressive correlated imaging, the fewer number of measurements is needed to obtain the reconstructed images.(2)It provides a wider application extent since imaging does not depend on the background.(3)For the regular moving objects, their locations and images can be recognized on the basis of their trajectory. For “lena” image with the size of 128 * 128, the binary images can be recovered under the compression ratio of 0.06. Subsequently, a clear image can be obtained with a compression ratio of 0.05 in the estimation range of moving objects.