| The panoramic video is a function that a plurality of cameras collect images,and after a series of processing such as camera calibration,splicing,blending,and encoding and transmission,the user can view the video wider than a single camera at the client side.In recent years,panoramic video has become a popular application in the field of virtual reality(VR)and security,mainly due to its better visual experience and greater amount of information.The single frame transmission of the panoramic camera is very simple for the common equipment,but the real-time video transmission needs to consider its huge data volume and huge processing burden.Among them,the real-time video splicing needs to consider the processing capability of the host computer.With regard to the current processor performance,it is not yet possible to automatically solve the splicing parameters on a frame-by-frame basis,and the splicing parameters that are calibrated in advance need to be processed persistently.Secondly,the panoramic video is composed of a single frame with a very high number of pixels,and it needs to be appropriately scaled according to the actual parsing quality of the camera.On the one hand,it reduces the processor load to a certain extent,and on the other hand,reduces the performance threshold for real-time transmission.At present,most of the schemes for implementing in-camera panoramic camera splicing are based on FPGAs and dedicated image processors(ISPs).However,the panoramic camera itself has a fast technology update and has a short product life cycle.If you take the customized route,it will inevitably lead to higher research and development difficulties and R&D.cost.Currently,the X86 platform already has low-power and low-cost solutions for handheld devices.If you can use the powerful instruction set and GPU resources of the x86 platform to achieve the same effects of FPGAs and ISPs,the economic value is extremely high.This topic observes the perspective of the panoramic camera splicing technology is relatively new,and the camera software and hardware system design is carried out from the perspectives of low cost,low power consumption,and real-time internal processing.Based on the theoretical research,this topic carries out a great deal of algorithm research,hardware selection and performance optimization.Due to space limitations,it is interspersed in various chapters as non-key points.This topic observes the existence of panoramic camera products on the market,and believes that the multi-channel camera's real-time in-camera splicing is costly and the calibration function is poor.This project compares and studies high-quality,high-computation image mosaics,and compares them with high-speed,low-quality algorithms.Finally,a robust splicing system based on GMS and ORB algorithms is selected as the An automatic stitching algorithm for calibration parameters is provided during stitching.Afterwards,the camera model is studied,and the methods for eliminating the distortion are compared and studied.The non-ideal condition of the fisheye lens is combined with the OpenCV library to realize the high-powered camera distortion correction function.Finally,a network panoramic camera scheme based on RTSP streaming is implemented.Subsequently,through the optimization of the performance of the global pipeline and the performance of the flow,a short and effective analysis is made and the solution is proposed.The focus of this article is to combine the theory and the theory and practical means of realizing the low-cost,low-power real-time in-machine splicing scheme.It aims to provide new ideas for the development of x86 low-power vision systems. |
PDF Full Text Request