Data-Mining-Based Computer Vision Analytics for Automated Helicopter Flight State Inference

摘要

The National Transportation Safety Board recently emphasized the importance of flight data monitoring as a tool to improve the safety and performance of helicopter operations. However, due to the high cost of using necessary equipment such as the flight data recorder, it has been difficult for the general aviation industry to participate in flight data monitoring programs by collecting flight data. To alleviate this problem, a possible workaround to infer the helicopter state information is to mount an inexpensive camera in the cockpit. In this paper, an image processing algorithm based on a density-based spatial clustering of applications with noise is proposed to accurately and efficiently estimate a helicopter's attitude, such as the bank angle, and read gauges in a flight instrument panel. The proposed algorithm initially uses the average motion energy technique to detect objects of interest from the successive image frames taken by a camera installed in a cockpit. Then, a data-mining algorithm called density-based spatial clustering of applications with noise clustering is used in the proposed algorithm to improve the performance of retrieving the flight state information through effective detection of a horizon line in the outside view and reading the gauge needles in the indicators. Finally, the proposed algorithm is successfully applied to a set of simulated video data generated by an X-Plane flight simulator, and real video data are taken in Sikorsky S-76 helicopters in order to demonstrate its desired performance.