| With the popularization of monitoring equipment,various enterprises have accumulated a large amount of video and image data,which has important reference value for understanding the situation of personnel in the enterprise.The detection of people in the monitoring data by the computer can quickly and effectively obtain the situation of the people in the target scene,which is of great help to the public safety in public places and the reasonable allocation of corporate resources.Object detection is an important content in the field of machine vision.This technology is widely used in monitoring,intelligent transportation,medical image analysis,marine ships,drones,new retailing,human-computer natural interaction,and other fields.The excellent performance of the current object detection model on public data sets can demonstrate that the model has strong generalization capabilities.However,in specific scenarios,performing object detection tasks on specific objects requires further improvement and fine-tuning of the network model to achieve better detection results.This paper builds an experimental data set based on the company's monitor video,improves the object detection algorithm,designs and implements a deep learning-based people detection system.It mainly includes the following research points:1.Construction of experimental data set:This paper collects the original video data,extracts the image frames containing people,and uses the image annotation tool to annotate the person information in the image,and processes the annotated data into general COCO data Set format to obtain the experimental data set.2.Improvement of the basic network of the object detection model:In the experimental data set constructed,due to the influence of the perspective and distance,the detection object has various sizes.In this paper,a convolutional neural network model based on Faster RCNN context information fusion is proposed.For multi-size detection objects,a deeper layer and better-effect basic network structure is used to achieve better detection.3.Improved loss function of object detection model and non-maximum suppression algorithm:Detection of object occlusion is also very common in data sets.In this paper,the loss function of the object detection model is improved by introducing a repulsive factor method,and the non-maximum suppression method of the object detection algorithm is improved to further improve the detection effect of the people detection model.4.Implementation of the people detection system:Based on the improved people detection model,in order to improve the efficiency of people detection processing for monitor video,this topic is based on the Django web application framework and uses Python to complete the development of the people detection system. |
