Research On Multi-source Video Registration Algorithm Based On DSP

With the rapid development of modern information and technology,people get more and more accesses to image information.Video and image contain a lot of intuitive information and data,and occupy an important position in people's access to information.With the growing maturity of infrared and visible video camera products,infrared video can directly reflect the dynamic distribution of radiation energy in the scene.It is widely used in the field of video surveillance under the condition of haze or low night illumination,but the image resolution is low,which is not conducive to human eye observation.Visible and video targets have high dynamic resolution and high resolution,but are vulnerable to weather and lighting conditions.If we can match the two video source images and take the advantages of each image,we can greatly improve the target observation ability.Therefore,we study the technology of infrared and visible video image registration.In the research of the video registration algorithm,according to the classical SIFT and SURF feature point detection algorithm theory,the implementation process and basic algorithm steps are analyzed.Then the simulation results of the classic algorithm are obtained by the experiment simulation and the advantages and disadvantages are analyzed.According to the idea of classical algorithm registration,the ORB feature point matching algorithm theory is analyzed,and the simulation results are also obtained.According to the comparison with the classical algorithm,the advantages and disadvantages of the algorithm are analyzed,and the G-ORB image matching algorithm based on the angle constraint of the feature point is proposed.The algorithm makes two improvements on the original ORB algorithm: constructing the Gauss Pyramid scale space of the picture and increasing the feature points.Detection scale space invariance;based on the problem of fast speed but high mismatch rate of the original ORB algorithm,the error matching point elimination method of feature point angle constraint is proposed for coarse purification.Under the same simulation conditions,it is compared with classical SIFT,SURF and original ORB algorithm.Then,we use RANSAC algorithm to do the refinement and determine the registration parameters,and get the best transformation matrix.Finally,we use bilinear interpolation method to complete the registration.The simulation results show that the registration speed is greatly improved when the registration accuracy is a little less than the classical SIFT algorithm,and it can meet the speed requirements of the registration of video images.In the hardware implementation of the G-ORB image matching algorithm based on the characteristic point direction angle constraint,in view of the characteristics of the performance requirements of the video processing,this topic uses the TMS320DM642 processor of TI company to complete the configuration of hardware peripherals and the design of each module in the software platform: the dual channel video acquisition module,the video storage module,the video registration module,the video display module and the process.The sequence of FLASH is burnt.The video registration module mainly completes the transplantation and engineering optimization of the improved algorithm.Because of the attribute of the binary descriptor of the G-ORB image matching algorithm of the feature point angle constraint,the share of the DSP on chip is far less than the SIFT and SURF algorithm,thus the algorithm can be transplanted to the DSP embedded platform.The experimental results show that the video registration effect is good and the dynamic information is fluent.