An Intruder Self-detection And Recognition Algorithm Based On Airborne Image

In recent years,the widespread use of unmanned aircraft has made the national airspace crowded,threatening the safety of UAVs and manned aircraft.To ensure the safety of aircraft in national airspace,how to integrate UAVs into the national airspace has become urgent problem to be solved.Sense and avoid technology of UAVs is a key technology to integrate UAVs into the national airspace.The technology is mainly divided into two parts: sense and avoid.Sense is a precondition to avoid,and an important part of the entire sense and avoid system.Therefore,the research on the intruder self-detection based on UAVs is significant.This paper mainly studies on the intruder self-detection and recognition algorithms based on airborne image of UAVs.First of all,this paper analyzes the common sensors in the sense and avoid system of UAVs.After considering the characteristics of SWAP(size,weight,energy power),load,sense distance and autonomy,a set of technical schemes of self-detection and recognition of intruder based on visual sensor is designed.The technical schemes is mainly divided into two modules: intruder detection module and intruder recognition module.Secondly,we study a kind of intruder detection algorithm based on edge detection.Edge images of airborne images is obtained by the Structured Forests for Fast Edge Detection method;the fully closed outline areas are marked with boxes based on the Edge-Boxes method,and the areas marked with boxes are considered as proposals that may contain an intruder;Finally,the proposals are scored,and then a set of optimal parameters is designed to filter the bad proposals to get the final region of interest.The simulation results show that the intruder detection algorithm based on edge detection can effectively detect the intruder under many different weather conditions.Compared with the intruder detection algorithm based on the Spectral Residual method,experiments show that the intruder detection algorithm based on Edge-Boxes method has higher recall rate and better detection performance.Thirdly,we study the algorithm of intruder recognition based on sparse coding and Spatial Pyramid Framework.Based on the characteristics of intruder with a complex background,the over-complete dictionary training is accomplished by using the commonly low-level feature;and then the sparse coding of the low-level feature is combined with the spatial pyramid framework to obtain more abstract sparse features;finally,the recognition of intruder candidate areas is accomplished by using linear support vector machine(SVM).The simulation results show that the SIFT-based intruder recognition algorithm is more robust and can effectively recognize the intruder candidate areas under different weather conditions.Finally,in order to verify the effectiveness and robustness of the intruder self-detection and recognition algorithm based on airborne image,an experimental platform was set up to validate the experiments in three aspects: different backgrounds,different illumination conditions and different numbers of the intruders.Experimental results show that the proposed algorithm can effectively detect and recognize the intruder with great robustness under different backgrounds,illumination conditions and the number of intruder.