Real-time GPU Accelerated Feature Point Tracker
Estimation of motion in image sequences is an important task in a wide range of applications. Tracking of feature points is a well-known and reliable technique to obtain motion information of salient points in video. In many applications this information needs to be computed in real-time, which normally limits the image resolution and the number of feature points that can be processed. Our GPU Accelerated Feature Point Tracker uses the enormous performance of graphics processors (GPUs) to reach a significant speedup. Due to the low cost of standard graphics boards this is also a very cost effective solution. Our software library implements the widely used KLT tracking algorithm on GPU, using NVIDIA CUDA . It tracks up to 10,000 feature points in real-time on full HD resolution.
The algorithm detects a set of salient feature points in every frame and adds it to the already existing set of feature points. Each feature point is then tracked to in the subsequent frame by minimizing a dissimilarity measure. The algorithm is deployed as a C++ software library, which can be easily integrated into applications. CUDA specific processing is entirely wrapped so that no CUDA knowledge is required for using the library.