Research And Implementation Of Construction Site Personnel Type Recognition Method

The construction of the power system is an important production and living project.It is necessary to supervise the personnel on the construction site to observe the safety production regulations at all times during the operation process to ensure the normal order of the construction.In order to achieve intelligent supervision of the construction site,people often use image processing algorithms in the field of target detection and target recognition to complete the positioning and identification of personnel on the construction site to achieve the supervision function,but common algorithms in this specific application scenario will be affected by the equipment conditional constraints and environmental interference factors make it difficult to run effectively.The widely used deep learning algorithms are also difficult to apply due to the particularity of the application background,which limits the application effect of the algorithm.In this paper,based on the characteristics of the application scenario,the detection and identification methods of construction site personnel are studied and implemented.The algorithm mainly combines visual attention mechanism,image color features,lightweight neural network and feature similarity matching related content.It has certain practical value to meet the application requirements when the image definition is low or the computing resources are limited.This article mainly introduces the algorithm of personnel target detection and personnel type recognition in special scenes such as construction site.The specific research work and results are as follows:(1)By studying the human visual attention mechanism and the bottom-up visual attention model,a human target detection algorithm based on prominent color features for regional positioning and combining with lightweight neural network is proposed.This algorithm uses the color information of the worker’s shirt image to guide the human significant region positioning and adopts.The human body detection model based on MobileNet-SSD network performs human body object detection on the full scene image and the distinctive color area image of the construction site.(2)The algorithm has many application sites and the input image resolution is low.Accurate face recognition cannot be implemented,and the on-site worker information can only be determined during the station shift stage of the day.The clothing characteristics of workers will change with the changes of the site and working day.As a result,it is impossible to use the deep learning method to train a long-term effective recognition model that is suitable for all venues,but the workers wear clothing with significant color features and hard hat features in the venue,and the clothing features are fixed during the identification process every working day.Combining this feature,this paper proposes a personnel type recognition method.The method achieved by extracting personnel safety helmet and dress color information to form a feature vector,and matching with the personnel database feature information obtained during the station shift stage of the day to determine the personnel category attribute.Established a safety violation rule judgment mechanism and integrated personnel target detection and identification methods to achieve the maintenance and supervision of the safety order on the construction site.(3)The dress color of electric power construction personnel is quite different from the surrounding scene environment.The dress color feature is an important symbol representing the attributes of personnel.It is necessary to accurately obtain this feature to determine the attributes of personnel.In this paper,based on the characteristics of HSI color space and Kmeans clustering algorithm,an improved color feature extraction algorithm is proposed,which can effectively extract the main color information of the image and provide a feature basis for the target detection and recognition algorithm.