Design And Realization On Aerial Video Based Target Detection System

Recently, Unmanned Aerial Vehicle Image System is widely used in reconnaissance and surveillance since it is low cost, high flexibility and user-friendly operation. This paper focuses on the design of the aerial image based object detection system which is specific to the real-time detection of the motion target under complex background and the road crossing. We also design a software system to validate the functions of our detection system. This paper adopts the scale-invariant feature transform (SIFT) to do the image registration to detect the motion target. The road crossing detection is based on the Hough line detection combined with the shape constraint. The motion target and road crossing detection are of great strategic value in aerial image and aerial video system, and our designed system is of great theoretical and application value in this field.The main work and contributions of this thesis are as follows:This thesis first analyzes the research background and significance of the unmanned aerial vehicle image system, and summarizes the related work and existing problems of the motion target and crossing detection, as well as the related fundamental knowledge.This thesis makes clear explanation of the SIFT algorithm, including its evolution and application. We also analyze the mis-matching of the feature points and high computation due to the complexity of the algorithm. To solve the above problems, this paper proposes an improved SIFT algorithm. The experiments show the efficiency of our improved algorithm.This thesis proposes a Hough line detection based road crossing detection algorithm and gives a complete framework for algorithm optimization.This thesis proposes a full solution to the unmanned aerial vehicle image based object detection system, elaborating the framework of the whole system and illustrating the applied software design of the system. Finally, the experiment demonstrates the efficiency of the solution.