Study On Moving Target Detection Algorithm Mobile Platform

Along with the rapid development of the growing popularity of smart handheld devices and networking, the moving target detection on mobile platforms become challenging issues in the presence of a strong parallax. Based on projective geometry, a relationship between the moving cameras in the presence of a strong parallax is analyzed, and unique constraints are proposed based on surface homography model. Generally, previous works focus on Planar+Parallax or simple geometric constraints such as fundamental matrix, but the low detection rate caused by degradation cannot be solved.Under the relevant work for moving targets detection on mobile platforms, this paper first introduces the development and application of moving target detection, and then discuss the causes of the projective parallax on a moving camera, and selects the projective parallax definition and size metrics that is easy to operate, and describes several traditional moving target detection algorithm and its advantages and disadvantages, points out that the moving target detection method based on a mobile platform in the presence of strong parallax is the inevitable trend of moving target detection method development. Surface Homography Model is described in detail in this paper, the most promising method is the algorithm of multi-view constraints based on the study and compare of the traditional algorithm based on optical flow constraints, and an improved adaptive algorithm based on multi-filters is proposed. In the improved algorithm, an adaptive framework with modeling, learning and detecting is introduced to solve the tricks of multi-vies and complex scenes, which is also instructive in general algorithms, such as target tracking, detection and identification. In the experimental part, by the actual image sequences in different cases to prove that the improved multi-view filters of the surface homography model can achieve a better detection result than traditional constraints in the presence of strong parallax. And the results show that this model can increase the performance such as detection rate and repeatable detection area by learning the motion of cameras efficiently, and be practical on moving target detection with a moving camera with a strong parallax.