基于可变形部件模型的安全头盔佩戴检测

摘要

Safety helmet wearing detection through video monitoring system is extremely meaningful for safety production ma-nagement.However,the present detection algorithms have too many application restrictions to meet the needs of various occa-sions.Therefore,this paper developed an improved algorithm to solve the problem.It proposed a modified deformable part model (DPM)by a combination of histogram of block-based local binary pattern (HOB-LBP),histogram of oriented gradient (HOG)and color features.With the usage of support vector machine for training and detecting,this algorithm effectively di-minished the loss of information compared with the original DPMwhich only included the HOG feature.According to the ex-periment on the safety helmet database,the average precision is increased by 7.2% and reached 86.7%,meeting the appli-cation requirements basically.%利用视频监控系统进行安全头盔佩戴检测，对于安全生产有着重要意义。已有的安全头盔佩戴检测算法有较多的应用场景条件限制，难以同时满足不同场景需求。针对这一问题，提出了一种应用于安全头盔佩戴检测的算法。该方法依托可变形部件模型，提出了基于块的局部二值模式直方图，与梯度方向直方图和颜色特征共同组成特征向量，使用支持向量机进行训练和检测，利用了单一使用梯度方向直方图作为特征时所损失的有效信息。实验结果表明，该方法优于原可变形部件模型，在安全头盔测试集上的平均检测率提高了7．2％，达到86．7％，已接近应用要求。