Image Target Detection And Analysis Based On Visual Computing

With the popularity of deep learning algorithms,the research of visual computing of images has attracted more and more resear-chers' attention.The research on various image analysis carried out in this series is endless.As a populous country in the world,China has a very high population density,and because of the contingency and unpredictability of human activities,there will be a large number of uncontrollable factors leading to accidents of one kind or another,which will create unnecessary danger wherever people are in,whether in daily life or in work and study.With the popularity of cameras,surveillance networks are ubiquitous,so situational understanding analysis based on visual computing has become a very necessary choice.This paper mainly proposes a target detection algorithm for situational understanding problem and a specific behavior detection algorithm based on pose estimation and its system function implementation.In the scene understanding experiment,the existing target algorithm is improved.The algorithm adjusts the bounding box by anchoring fi-ame technology,extracts more accurate features,designs different convolutional neural network models and different loss functions to complete training in different situations and building a database of different characteristics for training.In the behavioral detection experiment,the SSD target detection algorithm and the AlphaPose algorithm are used to complete the human pose estimation training.Then,based on the human body's posture estimation,the behavior detection analysis of the specific behavior-fall behavior was carried out,and an abnormal monitoring system for the elderly was designed based on this algorithm.The target detection algorithm of the scene understanding experiment was trained and simulated with different network models and different loss functions for the three different datasets constructed.The behavioral detection experiment uses the AlphaPose algorithm and the designed based on specific behavior fall detection algorithm to train and simulate tvwo different types of datasets,and based on this,the system implementation for abnormal monitoring of the elderly is completed.The simulation results show that the target detection algorithm can meet the real-time high-precision scene understanding experiment.The fall detection algorithm based on pose estimation can also perform the real-time accuracy specific behavior detection experiment,and the constructed system also has a good implementation effect.