A method for precise detection and recognition of UAV low-altitude targets

Abstract

The invention discloses a method for precise detection and identification of low-altitude targets of unmanned aerial vehicles. According to fully convolutional networks (Fully Convolutional Networks, FCN), the precise detection and identification of low-altitude targets of unmanned aerial vehicles based on a scale estimation model is realized; the method is based on low-altitude target vehicles, Motorcycles, pedestrians on bicycles, and pedestrians have the characteristic of obvious scale range. The pixel scale of the target is calculated through the model, and the parameters of the anchor are calculated at the same time to improve the accuracy of recognition.

Description

technical field [0001] The invention belongs to the fields of computer vision and artificial intelligence. Specifically, it relates to an accurate detection and recognition method for low-altitude targets of an unmanned aerial vehicle, which is used to improve the detection and recognition accuracy and speed of low-altitude targets. [0002] technical background [0003] Object detection requires locating the location of the object and identifying the class of the object. At present, general target detection is mainly based on R-CNN, Faster-RCNN, R-FCN, and R-FCN-3000 frameworks. The core idea of ​​this series of frameworks is to select candidate frames on the image and learn through CNN (Convolutional Neural Network). The R-CNN framework has achieved an average accuracy (mean Average Precision, mAP) of 58.5% in the VOC2007 data set test, and the Faster-RCNN framework can reach 73% in the VOC2007 mAP, and the NVIDIA Tesla K40 GPU speed can reach 5fps ( The number of frames...

AI Technical Summary

Problems solved by technology

Although the speed is faster than the previous framework, the accuracy for complex datasets drops significantly

Images