Velocity Estimation Using an Improved Image Local Feature Representation with Enhanced SIFT Descriptor for Mobility

Velocity measurement based on visual data received from a camera has garnered tremendous attention in the fields of computer vision and image processing. The camera resolution, view angle, and depth stability affect the performance of velocity measurement frameworks. Various descriptors have been developed by researchers; however, they are severely prone to photometric and geometric variations within the image. In this research, we propose an enhanced SIFT local descriptor to extract features from the local structures within the image. Based on feature matching between consecutive frames of the video, we estimated the velocity of a UAV flying at a uniform height above the ground. The proposed descriptor resulted in a more accurate velocity estimate than many state-of-the-art descriptors available in the literature.