Reading detection of needle-type instrument in a noisy environment using computer vision-based algorithms

Abstract

This study investigated the use of computer vision–based algorithms for detecting needle-type instrument readings in the cockpit of an aerial vehicle. A flight data recorder plays a crucial role in aviation safety investigations and flight operation review. In practice, not all aerial vehicles are equipped with flight data recorders, and this poses problems in the retrieval of instrument readings during investigations. Installing a lightweight recorder such as a camera in the cockpit to record the instrument panel is a solution to the mentioned problem. Although recorded flight data can be retrieved through human inspection, computer vision–based algorithms enable more rapid and efficient detection. Accordingly, this study developed two computer vision–based algorithms operated in both of the grayscale color space and the value layer of the hue–saturation–value color space. Performance of the four combinations is then compared and the best combination of algorithm along with operation space is suggested. The airspeed meter of a Bell 206 helicopter was selected to test the proposed detection algorithm in this study. GPS data and human inspection results were used as references. Herein, experimental results are presented, performance of algorithms is discussed, and conclusions are provided. This study contributes to aviation safety investigations and flight operation review involving aerial vehicles that are not equipped with flight data recorders.

Appendices

Appendix A: Abbreviation

The following abbreviations are used in this manuscript:

CVR:

Cockpit voice recorder

DBSCAN:

Density-based spatial clustering of applications with noise

FDR:

Flight data recorder

GPS:

Global Positioning System

HSV:

Hue-saturation-value

ICAO:

International Civil Aviation Organization

RGB:

Red-green-blue

STD:

Standard deviation

TTSB:

Taiwan Transportation Safety Board

Appendix B: Pseudo code

1.1 B.1 Transformation of color space

1.2 B.2 Binarization and edge detection

Suppose that the thresholds for binarization are $(T_{\min \limits },T_{\max \limits })$. The edge detection employs Matlab function edge() directly. The pesudo code is presented as follows:

1.3 B.3 Build of needle mask

1.4 B.4 Needle angle detection using mask

1.5 B.5 Identification of needle from background

The biggest object in the binary image is identified by Matlab function bwconncomp(). Suppose that the image is B.jpg. The Matlab code is presented as follows:

function [NDL,biggest] = NDL_idf(Bunit8) Bunit8 = impread('B.jpg'); Bdble = double(Bunit8); Bnml = Bdble/255; Bcnd = bwconncomp(Bnml); neighborhood. index = 1:Bcnd.NumObjects; numPixels = cellfun(@numel,B.PixelIdxList); Identify the index of the biggest object [biggest,idx] = max(numPixels); index(idx) = []; Bnml(Bcnd.PixelIdxList{index(i)}) = 0; objects end NDL = Bnml;

1.6 B.6: Needle angle detection using image moment