Multi-directional unmanned aerial vehicle aerial photography vehicle detection method based on attention mechanism

Abstract

The invention discloses a multi-directional unmanned aerial vehicle aerial photography vehicle detection method based on an attention mechanism. A feature pyramid network, namely an RPN network, fused with the attention mechanism is designed; according to vehicle characteristics, an RPN network is designed and direction attributes are added; a Faster R-CNN is improved, and the scheme of the two steps is added; and the vehicle is detected to obtain a result. According to the invention, in the Faster R-CNN design, through adding an attention mechanism, each channel and region of the FPN can obtain an adaptive weight, and vehicles in multiple directions can be accurately detected, so that the vehicles can be better counted, and the situations of missing detection and false detection are reduced.

Description

technical field [0001] The invention belongs to the field of target detection of unmanned aerial vehicle images, and relates to the current advanced target detection algorithm, attention mechanism and multi-directional vehicle detection module of aerial images based on FasterR-CNN. Background technique [0002] With its high mobility, rapid deployment and large-scale surveillance, drones play an important role in traffic management, parking lot management, urban planning, emergency rescue, etc. In recent years, with the increase of vehicle ownership, traffic problems have become increasingly prominent. With the help of vehicle detection, traffic flow calculation and traffic congestion prediction can be performed, so as to better monitor and manage traffic. All of this requires the ability to accurately detect all vehicles. [0003] In recent years, with the improvement of graphics processing technology, deep learning and artificial intelligence have achieved great success ...

AI Technical Summary

Problems solved by technology

The existence of non-vehicle linear objects, such as power units and air-conditioning units on top of buildings, will complicate the task and cause many false alarms, and the target objects in mid- and low-altitude images are relatively small, generally only 15-20 pixels, so it is necessary to Human-machine photography of vehicles for accurate detection is a big challenge

Images