Research On Helmet Detection Algorithm For Industrial Site

With the wide application of image processing in various fields of life,the industrial field has gradually proposed to apply image recognition to industrial production and the helmet is the most widely used head protection tool in the construction area,if you don't wear a helmet,will cause greater security risks.At present,the helmet detection algorithm has problems such as large external interference and low detection accuracy,and most of them are to determine whether the worker wears a helmet by recognizing the shape and color of the safety helmet.So,this paper has conducted related research on such issues,and the main results achieved are summarized as follows:(1)We propose a face detection method based on skin color segmentation and EHSW-Adaboost algorithm to locate the face region.Firstly,to obtain candidate face regions by using the Gaussian skin color model in YCbCr color space to detect the images.Then,improving the Adaboost algorithm to deal with the missing detection problems in the algorithm and the degradation phenomenon that occurs when training the classifier.That is,to make the trained classifier can detect the tilted face by extending the Haar-like features,and adjust the update strategy of sample weights to avoid over-distortion of sample weights in the training process.Finally,using the EHSW-Adaboost algorithm to confirm the candidate's face area,and then locate the helmet wearing area for subsequent safety helmet detection.(2)A safety helmet detection method based on SW-Adaboost is proposed.Firstly,the edge feature and linear feature which can reflect the information of the helmet are selected from the haar-like feature template.Then,using the Adaboost algorithm based on sample weight update to detect the helmet,the algorithm changes the sample weight update strategy.In this strategy,an update threshold is defined by each iteration,and the adjustment of the sample weight is determined based on the classification results of this round of samples,the threshold and the relationship between the sample weights,so as to prevent the excessive increase of the sample weights in the learning process.Finally,the exact helmet area is obtained.The above results can not only enrich the method system based on image helmet detection,but also effectively support the extended application of image recognition technology in the industrial field.