| In recent years,wireless multimedia sensor networks(WMSN)have attracted wide attention in the field of monitoring and control.It is difficult to meet the requirements of rapid deployment to the WMSN nodes,so it is necessary to enhance the range of perception,and study the WMSN nodes with 360 degrees of omni-directional sensing characteristics.Therefore,based on the requirement of omnidirectional vision of WMSN node,this thesis has studied the panorama stitching algorithm based on multiple wide-angle cameras and designed omni-directional visual perception system based on multi-core ARM embedded platform,which has positive scientific significance and application value.As the use of multiple wide-angle camera omni-directional vision,image distortion correction is the first problem in this thesis.In this thesis,Zhang Zhengyou calibration method is introduced for wide-angle camera calibration and the cameras' internal parameters are acquired.In addition,several different correction algorithms are studied and analyzed,then the actual results of different calibration algorithms are contrasted.Finally,the inverse division correction model is adopted to achieve image correction,and it has a high robustness and perfect correction effect.Before the image registration,the SIFT and SURF features of the images are extracted.By analyzing and comparing the detection results and matching results of the two algorithms,the SIFT algorithm with better stability is adopted in this thesis.The feature vector matching algorithm with higher real-time and RANSAC algorithm are used to match of key points.In order to obtain better matching results,normalization algorithm is also used.For image registration,the global homography method and SPHP algorithm are studied and analyzed.Contrast to the global homography method,SPHP algorithm performance is excellent.However,when parallax is large,there will still be registration error with SPHP algorithm.Aiming at this special situation,this thesis proposes a new image registration method based on motion detection and SPHP algorithm,which could improve the accuracy of image registration under the condition of large parallax.For the existence of joint problems after the adjacent image mosaic,this thesis discusses and analyzes several algorithms of image fusion.By comparing the character of real-time and the effects of different methods,the weighted average method is used for image fusion.When the brightness difference between adjacent images is large,the brightness difference of the fused image is still very obvious.To solve this problem,this thesis analyzes and compares the method of histogram equalization algorithm and luminance equalization method based on overlapped,and put forward a new brightness balanced method based on the matched feature points,and the new method effectively solve the problem of uneven brightness of image.In the last part of this thesis,the hardware configuration of i.MX6Q multi-core processor and the performance of PCIe acquisition card are introduced.In order to get more effective performance,the embedded Linux kernel clipping was performed on the computer platform and some unnecessary drivers were removed.To further improve the operating efficiency of the system,this thesis uses a number of hardware co-processor.In the algorithm,coordinate mapping method is adopted to simplify the processing flow and reduce the unnecessary calculation.The multi-threading technology is used to capture and process images simultaneously.Finally the omni-directional visual perception system based on multi-core ARM embedded platform is realized... |
PDF Full Text Request