A Comparative Study of Machine Learning Classification for Color-Based Safety Vest Detection on Construction-Site Images

Detecting the safety vests is an important foundation for various applications in safety management and productivity measurement. The fluorescent yellow-green color and fluorescent orange-red color of safety vests are generally considered as the most distinctive colors which represent workers in construction-site images. The objective of this study is to provide an evaluation of the safety vest detection using color information in construction-site images. The data sets of two colors of safety vests and the background were generated and used in this study. A comparative analysis of combinations of five color spaces (RGB, nRGB, HSV, Lab, and YCbCr) and six classifiers (ANN, C4.5, KNN, LR, NB, and SVM) was conducted. The performance of each combination was assessed in terms of the precision, recall, and F-measure. Moreover, an evaluation of the effects of color space conversion and the absence of luminance components on the detection performance was conducted. The comparison results showed that C4.5 classifier combined with YCbCr and SVM classifier combined with Lab, respectively, outperformed other combinations on each data set of safety vest colors. Furthermore, RGB color space transformation into non-RGB color spaces enhanced the classification performance. The evaluation also showed that the removal of luminance components did not help to improve the performance.