Improved Helmet Wearing Detection Method Based on YOLOv3

In order to monitor the wearing of safety helmet in dangerous workplace in real-time and ensure the safety of production, this paper proposes a method based on the yolov3 algorithm to detect the wearing of safety helmet. Firstly, K-means algorithm is used

  • PDF / 2,948,482 Bytes
  • 12 Pages / 439.37 x 666.142 pts Page_size
  • 24 Downloads / 219 Views

DOWNLOAD

REPORT


2

1 School of Computer Science and Information Security, Guilin University of Electronic Technology, Guilin 541004, China [email protected] Guangxi Construction Industry Mostly Lease Co. Ltd., Guilin 530000, China [email protected]

Abstract. In order to monitor the wearing of safety helmet in dangerous workplace in real-time and ensure the safety of production, this paper proposes a method based on the yolov3 algorithm to detect the wearing of safety helmet. Firstly, K-means algorithm is used to cluster the target boxes on the self-made data set, so that the prediction boxes can fit the target better that in the data set. At the same time, the network is pre-trained on the voc2007 data set to make the model parameters more accurate and reduce the training time. Secondly, multi-scale feature extraction and multi anchor box mechanism are used to improve the accuracy of small object detection. Finally, optimizing the non-maximum suppression (NMS) algorithm with Gaussian function that can improve the detection accuracy of the occluded target. Experimental results show that the algorithm has better detection effect while meeting the real-time helmet wearing detection.

Keywords: Helmet detection detection · YOLOv3

1

· Non-maximum suppression · Object

Introduction

In many complex production environments, such as construction sites, mining and other places, there are various risk factors threatening personal safety. As the head is the most critical part of the human body and the most vulnerable to fatal injury, wearing a helmet in such workplace is an effective guarantee for the life safety of personnel in the production area. Therefore, monitoring whether the personnel in the production area wear safety helmet in real-time is an important means to ensure the safety of production. This work is supported in part by the National Natural Science Foundation of China (Nos. 61702129, 61772149, and U1701267), Innovation Project of Guet Graduate Education (No. 2019YCXS050), and Guangxi Natural Science Foundation (Nos. 2018GNSFAA138132, AD18216004, and AD18281079). c Springer Nature Switzerland AG 2020  X. Sun et al. (Eds.): ICAIS 2020, LNCS 12239, pp. 670–681, 2020. https://doi.org/10.1007/978-3-030-57884-8_59

Improved Helmet Wearing Detection Method Based on YOLOv3

671

Traditional helmet detection methods are primarily implemented by handdesigning features. Jia, et al. proposed a feature vector composed of a blockbased local binary histogram based on the variability component model and the gradient direction histogram and color features, using the support vector machine (SVM) to achieve the helmet detection [7]. Liu, et al. used a skin color detection method to locate the face region, using the Hu matrix of the face and above region as the feature vector of the image, training the model with BP neural network and SVM to implement helmet detection [9]. Feng, et al. used the mixed Gaussian model to detect the foreground, judged whether it belongs to the human body through the connected region, and finally positioned the hum

Data Loading...