Research Of Electronic Image Stabilization Technology Based On DSP

As a content-based image stabilization method, electronic image stabilization (EIS) hasadvantages of low power consumption, small volume, convenient installation and debugging,high precision. It doesn’t need to pre-get the motion model or mechanical movement of thecamera, only needs to calculate between the adjacent frames, the motion vectors are got, soEIS is widely used in many kinds of video monitor and surveillance situation. The main workof the thesis is studying EIS algorithms, realizing the embedded EIS system based on DSPand verifying the effectiveness of embedded EIS system by experiment.First, this thesis deeply analyses common algorithms such as gray projection algorithm,feature point matching, bit-plane matching and block matching. The main difference betweenthese common algorithms is the method which is used to calculate the motion vectors betweenthe adjacent frames. Based on the analysis of the algorithm principle and implementationmethod, this thesis verifies each of the common algorithms by simulation experiment,compares advantages, disadvantages and applicable situation of the common algorithms. Thisthesis verifies that block matching video stabilization algorithm has high-precisionstabilization, wide suitability and Faster computing rate provides theoretical and experimentalbasis for the new algorithm designed in this thesis.Second, this thesis puts forward a new adaptive block matching EIS algorithm toimprove the disadvantages that traditional block matching video stabilization algorithmscannot remove the influence of moving object completely and needs a preset threshold valueto decide similarity in processing procedure and a lot of calculation work. The new EISalgorithm searches the matching blocks between adjacent frames independently in twodimensions, and determines the optimal location adaptively without the preset similaritythreshold, and according to the optimal location, the algorithm removes the moving objectareas. This algorithm for three-step search algorithm was improved, reducing the number ofmatching algorithm, at the same time compared with diamond search method can effectivelyovercome the error of the local extreme value point. The result shows that the new EISalgorithm can stabilize the video that has large size dithering and exists some moving objectson the scene.Compared with the general PC platform, DSP platform has higher flexibility, smallervolume, lower power consumption, higher operation efficiency. This thesis uses a specialimage processing chip TMS320DM642that TI Company DaVinci series as hardware processing platform. DM642has Three configurable video port, a network interface, a PCIinterface, a16-bit general input/output port and EDMA controller. DM642is very suitable forDigital image processing algorithm.In order to realize the real-time capability of the proposed algorithm and reduce thevolume and power of the processing platform, this thesis transplants and optimizes theproposed algorithm for DSP embedded video processing platform. The result shows that onesingle frame processing time is3.52ms, which completely meets the need of standard videoframe rate.Finally, based on the DSP embedded video processing system, this thesis carried out theEIS experiments on many modes, such as different camera motion states, simple and complexbackground. Experiments’ result shows that the proposed algorithm can effectively stabilizethe two dimensions translational motion of the dithering video on the DSP embedded videoprocessing platform.