營建工地安全系統之工地安全帽及背心偵測

 

Helmet and Vest Detection for Construction Site Safety System

 

陳韋成

國立臺灣大學 工程科學及海洋工程學系

羅斯福路四段1

台北市10617大安區

 

丁肇隆

國立臺灣大學 工程科學及海洋工程學系

羅斯福路四段1

台北市10617大安區

 

張瑞益*

國立臺灣大學 工程科學及海洋工程學系

羅斯福路四段1

台北市10617大安區

*rayichang@ntu.edu.tw

 

摘要

營造業之工地事故職業災害發生比率,相較於其他行業更為嚴重。其中,工地人員未按規定配戴安全裝備,是造成嚴重傷害的主要原因之。為了避免悲劇發生,本研究藉由人體偵測及物件辨識技術,判定勞工進入工地時是否按規定著裝,如工地安全帽及工地背心,以達到自動化的安全檢查。本論文主要分為三大部分:影像前處理、特徵擷取及辨識。首先,以網路攝影機拍攝影像後,透過背景分離法擷取移動的前景影像,接著定位出工地安全帽及工地背心位置。最後,再抽取色調直方圖、飽和直方圖以及區域二元圖形(Local Binary Pattern, LBP)作為影像特徵,交由支持向量機(Support Vector Machine, SVM)進行分類。實驗結果顯示,本系統每次偵測之執行時間平均只有270毫秒,辨識工地安全帽及工地背心正確率分別為97%93%,可以有效降低人力負擔需求。

 

關鍵詞: 營建工地安全、圖形辨識、影像處理、安全帽偵測、背心偵測。


 

 

 

Abstract

The occurrence rate of severe occupational injury in construction industry is much higher than others. This high risk is primarily caused by the deficiency of the personal protective equipment. In this paper, we apply the techniques of body detection and object recognition to design an automatic checking system for examining whether construction workers are equipped safety helmets/vests or not. There are three parts in our system which including image preprocessing, feature extraction and pattern recognition. First, videos of workers are taken by an IP camera. Then, the images of workers would be extracted by background subtraction, and the positions of safety helmets and safety vests are located. At last, the support vector machine (SVM) is utilized to perform classification with the image features of hue histogram, saturation histogram and local binary pattern (LBP). The experiment results show the system could effectively recognize safety helmets and safety vests with the accuracies of 97% and 93%, respectively.

 

Keywords: Construction site safety, Pattern recognition, Image processing, Helmet detection, Vest detection.