Unmanned aerial vehicle real-time moving target classification and detection method based on vision

Abstract

The invention discloses an unmanned aerial vehicle real-time moving object classification and detection method based on vision. The target identification is carried out by means of an advanced technology of deep learning, an unmanned aerial vehicle can accurately identify similar targets in a video through an advanced YLOv2 algorithm, and is used for carrying out statistics and marking on the similar target objects to facilitate a user's use; and after the user selects a certain specific target object in the identified target object at a ground station, the unmanned aerial vehicle system usesan ORB algorithm to extract the characteristics, and then is continuously matched with the feature of the similar target objects extracted from each frame of video, and through the combination with the motion trail trend of the selected target, the final matched specific target and the position are comprehensively obtained. The specific target recognition provides guarantee for subsequent automatic tracking of targets, precise landing, and other automated functions. The method solves the real-time airborne target recognition and detection, including the recognition of the similar target objects and the specific target object, of the unmanned aerial vehicle in a complex dynamic background, and takes into account the requirements of speed and precision detection.

Description

technical field

[0001] The invention belongs to the fields of UAV image processing technology and computer vision, and in particular relates to a method for classifying and detecting real-time moving objects by UAVs based on vision and deep learning. Background technique

[0002] Real-time recognition of moving targets in complex dynamic backgrounds is a necessary step for UAVs to be fully autonomous. However, due to the particularity of the application platform, traditional detection methods such as segmentation-based and classification-based Because the region selection strategy based on the sliding window is not targeted, the time complexity is high, the window is redundant, and the robustness to the detection of the dynamic background is not high; and the deep learning based on R-CNN as the representative Combining the target detection framework of region proposal and CNN classification, because the application requirements of detection accuracy and detection speed canno...

Images