Detection of SIFTKeypointsin Spherical Omnidirectional View Sensor

AbstractThispaperproposed a methodtodetectobject/scenethroughScaleInvariantFeatureTransform(SIFT) keypointsforahybrid sensor systemcomprises of aperspective viewsensor and aspherical omnidirectional viewsensor.A referenceimageisobtained from the perspective view sensor and matching is attempted with another distorted image acquired from the omnidirectional camera.Theomnidirectional viewimageisfirst subjected todistortioncorrection(commonly termed “unwrapping”) using closed form mappingfunctions and thenSIFTkeypoints of the correctedimage areextracted and matched againstthereferenceimage's features. Experiment results show that the distortion correction produces acceptable performance of SIFT keypoint matching without modification to the classicSIFT algorithm