An unmanned aerial vehicle real-time detection method used in a dynamic environment

Abstract

The invention discloses an unmanned aerial vehicle real-time detection method used in a dynamic environment. The unmanned aerial vehicle real-time detection method comprises the steps of 1, establishing an Otsu-HSV color model; Step 2, inputting an image into the Otsu-HSV color model; using a BING algorithm to detect, and combining an overlapping area formula to eliminate a part of detection windows with large coincidence degrees; and step 3, combining an NMS algorithm of a linear weighting method, and replacing the original confidence score of some windows needing to be suppressed with relatively low confidence by using the NMS algorithm. According to the method, the HSV space and the maximum between-cluster variance method are combined, the overlapping area and the linear weighted scoreare introduced, the improved BING / NMS algorithm is provided, and finally, on the premise that the real-time performance is met, the detection accuracy and the optimal overlapping rate of the unmannedaerial vehicle are ensured, and the false detection rate and the missing detection rate are reduced.

Description

technical field [0001] The invention relates to an unmanned aerial vehicle detection method, in particular to a real-time detection method for an unmanned aerial vehicle in a dynamic environment. Background technique [0002] In modern air combat, drones have played an increasingly important role due to their small size, low cost, and high safety. Among them, the detection and recognition of enemy targets has always been a research hotspot in the industry. With the continuous development of aerospace technology, the battlefield environment has higher and higher requirements for real-time performance and accuracy. The method of manual interpretation of aerial image detection can no longer adapt to modern The needs of information warfare. If objects could be detected and recognized automatically, then truly autonomous drone flight would be feasible. [0003] Target detection based on RGB color space, when color separation is performed, it is prone to mis-separation, missing ...

AI Technical Summary

Problems solved by technology

UAV cluster flight is greatly affected by illumination and mutual occlusion, and the background is complex in a dynamic environment. In addition, when the traditional BING algorithm detects UAV clusters, the large detection bounding box will often completely cover the small detection bounding box. , resulting in some repeated candidate frames, resulting in false detection; when the UAV clusters occlude each other, when the overlapping area of ​​the two UAV candidate frames exceeds the threshold of the non-maximum value suppression algorithm, according to the traditional BING algorithm post-processing The non-maximum value suppression algorithm can only keep one UAV, resulting in missed detection. The above problems make it impossible to meet the real-time and accurate processing requirements of UAV detection.

Images