On Co-detection Of Low-altitude,Slow-speed,Small Targets

Target detection is a basic research task in the field of computer vision,which has been con-tinuously researched and successfully applied in many other fields.Reliable target detection is the premise of further disposal tasks,such as target tracking and analysis.As an important component of target detection field,the low-altitude,slow-speed,small-target detection has received a lot of attention due to its difficulties.The difficulties include complex low-altitude observation environ-ment,existing strong clutter and strong random noise,low signal-to-clutter ratio or signal-to-noise ratio,lack of relative movement,among others.In recent years,the co-detection methods have been introduced into the target detection field,which aim to make full use of the consistent appear-ance and the motion continuity of targets observed in multiple frames to form judgment evidence chain and give reliable detection results.This provides a new way for reliable detection of low-altitude,slow-speed,small targets.In this thesis,different observation scenarios for low-altitude,slow-speed,small targets are considered,focusing on complex background suppression,target in-formation accumulation and target association detection from the perspective of co-detection.The main contributions of this thesis work are as follows:1.Aiming at the problems associated with low signal-to-noise ratio of low-altitude,slow-speed,small-target detection,the influence of superposition projection on target information ac-cumulation is analyzed.First,the energy decay trend of the noise synthesis image is obtained by the least squares fitting method.Second,the energy change of the target synthesis image is ana-lyzed,and the analytic expression of energy attenuation of the target synthesis image is obtained under several overlapping conditions.Analysis shows that superposition projection can enhance the signal-to-noise ratio of small targets.2.Aiming at the problems associated with low-altitude,slow-speed,small-target detection in the stationary background,a co-detection method based on superposition projection,singular value decomposition and nearest neighbor association strategy is proposed.First,an overall target detection method is proposed based on the angle criterion,which can quickly and preliminarily evaluate the existence of the target.Second,based on the motion characteristics of the target,a target association model is proposed by using singular vector reconstruction and neighborhood correlation strategy.It can solve the detection problem caused by different signal-to-noise ratios or target loss.3.Aiming at the problems associated with low-altitude,slow-speed,small-target detection in the slowly changing background,a co-detection method is proposed based on temporal target de-tection model and graph association strategy.First,a three-phase suspicious target detection model is established based on temporal filtering,temporal target fusion and cross-product filtering.It can effectively suppress both slowly changing clutter and strong random noise.Second,according to the motion and appearance characteristics of the target,a target association model is obtained based on the graph association strategy.It can effectively suppress the highly-suspicious targets and false alarms.4.Aiming at the problems associated with low-altitude,slow-speed,small-target detection in the fast changing background,a co-detection method is proposed based on target reconstruction and spatio-temporal association strategy.First,dense and sparse target reconstruction models are proposed based on nonlinear weights and element-by-element weights,respectively,which can effectively separate the background and target information in single-frame images.According to the common characteristics of the targets in the spatio-temporal domain,a target association model is proposed based on single-frame position association and multi-frame trajectory correlation.It can effectively improve detection reliability and efficiency.