The Research Of Vehicle Detection In Aerial Imagery

Automatic target detection and recognition is a key technology of next-generation onboard photoelectric reconnaissance pods.Its main purpose is to detect and locate specific targets by utilizing visual information obtained by photoelectronic pods.Target detection and recognition under aerial scene is an important research branch in the field of computer vision and is of great significance both in research and practice for military and civil applications.Due to the relatively small target and variation of pose and complexity of background,the detection and recognition of typical target under aerial scene has always been a concern of many scholars.In this paper,aiming at the problem of vehicle detection in aerial scene,a airborne vehicle detection is designed and the vehicle detection algorithm based on aerial images is studied.The specific work includes:Firstly,the design requirements of the vehicle detection system are analyzed.In view of the shortcomings of the traditional design framework,and the system hardware architecture and software framework are designed.Secondly,the vehicle detection methods for aerial scenes are studied,and the general target detection framework is improved.The vehicle detection process is divided into two parts: the regions of interest extraction and the DPM vehicle detection model.Based on the analysis of vehicle target characteristics,a method of region of interest extraction based on visual saliency is studied.The saliency features based on frequency tuning are obtained,and the top-hat operator is used to extract the ROIs.The experimental results show that the method can effectively narrow the vehicle search range and improve the detection efficiency.In order to detect the vehicles,the aerial vehicle dataset is established.The vehicle detection algorithm based on the deformable parts model is studied.The key parameters of the model are adjusted and applied to the vehicle detection under aerial scene.Furthermore,aiming at the problem that the detection rate of the traditional method is not high enough,the Faster RCNN-based aerial vehicle detection algorithm is studied.The candidate area can be obtained through the region-proposal network RPN,and then the object is located in the image by Fast RCNN.The experimental results show that this method can achieve higher detection accuracy than the deformable parts model.Finally,the aerial vehicle detection system is implemented on the embedded platform,and the key functional modules are introduced.The Faster RCNN based vehicle detection algorithm was ported to the JETSON TX1 development kit to verify the effectiveness of the aerial vehicle detection system and lay the foundation for the functional integration of the airborne pods.