An event-driven asynchronous feature tracking method

Compared to traditional frame-based cameras, event cameras offer a few advantages due to their bio-inspired characteristic, including low latency and high dynamic range, as each pixel can independently respond to changes in brightness when generating event streams. However, the unconventional output of event cameras brings challenges to basic computer vision problems. Recent research in event-based feature tracking has focused on using information from event streams to achieve asynchronous tracking of feature positions. We proposed an improved feature tracking method with complementary information from both the event and frame-based cameras to update feature positions asynchronously. A nearest neighbor algorithm is used for data association to filter new feature points, and the optimization quality is evaluated by the cost function values. This method utilizes the detection of FAST corners on the RGB camera to initialize the missing objects or targets, which are points exhibiting significant local pixel brightness changes. The algorithm has been implemented in C++ and estimated using publicly available datasets. The experimental results indicate that our method boosts the feature age by 9%$\sim$24% while maintaining tracking accuracy compared to the Event-based Kanade-Lucas-Tomasi tracker (EKLT) method.