PROJECT TITLE :

Detection of Small Aerial Object Using Random Projection Feature With Region Clustering

ABSTRACT:

The detection of small aerial objects is an important component of a wide variety of computer vision tasks, such as remote sensing, early warning systems, and visual tracking, amongst others. Despite the fact that there are techniques for detecting moving objects that can achieve reasonable results when applied to normal-sized objects, these techniques are unable to differentiate small objects from the dynamic background. In order to find a solution to this problem, an innovative method for the precise detection of small aerial objects in a variety of environments has been proposed. At first, the block segmentation was implemented in order to cut down on the redundant nature of the frame information. During this time, a random projection feature known as RPF is being considered for use in characterizing blocks into feature vectors. After that, an estimation of the moving direction based on feature vectors is presented in order to measure the motions of the blocks and filter out the major directions. Finally, in order to extract pixelwise targets from the remaining moving direction blocks, variable search region clustering (VSRC), in conjunction with the color feature difference, was developed. The exhaustive experiments show that our approach outperforms the level of state-of-the-art methods upon the integrity of small aerial objects, particularly on the dynamic background and scale variation targets. This is especially true for the dynamic background and scale variation targets.