Research And Implementation Of The SIFT Stereo Matching Algorithm Based On FPGA

Scale-Invariant Feature Transform (SIFT) can be used to perform reliable matchingbetween different views of an object or scene. It is invariant to image scale and rotation,and a substantial range of affine distortion. It has a wide range of applications in thebinocular stereo vision. This paper studies how to realize the scale-invariant featuretransform algorithm in FPGA effectively.Firstly, the paper analyzes the basic principles and process of the scale-invariantfeature transform algorithm. According to the characteristics of this algorithm, aFPGA-based hardware/software co-processing solution is given. This design is based onTop-down modular structure, and achieves the Difference of Gaussian scale space byusing the Gaussian filter module, and computes the gradient magnitude and orientationof the keypoint by calling the CORDIC IP core in system, and obtains the local imagedescriptor by using the soft-core NiosⅡ embedded processor. The experiment on twoimages with a resolution of320x240in different views verifies that the design meetthe requirements.Considering the hardware resources of the platform, this system uses the pipelinedesign to reduce the utilization of resources greatly, while ensuring the processing rateof the system. The modular design facilitates the reuse of the module and makes thesystem easy to be expanded for higher resolution and greater matching range.