Video target detection method for aerial images of unmanned aerial vehicle

Abstract

The invention discloses a video target detection method for aerial images of an unmanned aerial vehicle. The method comprises the following steps: carrying out the data preprocessing of each patch (batch) of a training set, and realizing the multi-scale training; adding a feature pyramid and deformable convolution into the feature extraction network ResNeXt to enhance the ability of the feature extraction network; a plurality of detectors are cascaded, the first detector performs target classification and coordinate regression of a suggestion box output by an RPN (Regional Recommendation Network), and the next-level detector further processes the fine suggestion box output by the previous level until the last-level detector ends; and replacing the feature extraction network for retrainingto obtain two models, performing multi-scale test on the two models, and then performing multi-model fusion by adopting a softnms (non-maximum suppression) method to obtain a detection result. The method has high detection precision and strong robustness under the condition that the number of targets in aerial images of the unmanned aerial vehicle is large and the size is small.

Description

technical field [0001] The invention relates to the fields of image processing and computer vision, in particular to a video target detection method for drone aerial images. Background technique [0002] Today, drones or general-purpose drones equipped with cameras have been widely used in many fields such as agriculture, aerial photography, fast delivery, surveillance, etc. Different from conventional detection data sets, each picture of UAV aerial images contains hundreds of objects to be detected, and the annotation frames of pedestrians and distant objects are very small, and the occlusion between objects occurs frequently, which is difficult to detect. Object detection in man-machine aerial images brings difficulties and challenges. [0003] In recent years, computer vision has made major breakthroughs in image recognition, target detection, image segmentation and other fields through deep learning and other technologies. Video object detection is a basic task in comp...

AI Technical Summary

Problems solved by technology

The above deep learning algorithms are suitable for conventional scenes. For the case of multiple targets and small target sizes in drone aerial photography scenes, the detection accuracy is significantly reduced.

Images