Research Of Wearable Pre-warning Technology Based On Machine Vision

With the rapid development of embedded technology and video surveillance equipment,the comprehensive application of machine vision technology is becoming more and more widespread,especially security monitoring situation.This technology involves a comprehensive application of many technologies such as image processing and artificial intelligence.Public safety problems that frequently occurred in recent years have brought new challenges to monitor and pre-warning systems.Nowadays,monitoring and pre-warning systems are mostly based on PC platforms and have been widely used.With the development of embedded software and hardware technology,more and more image algorithms and sensors can be applied to embedded platforms.A hardware platform,starting from the overall planning of the wearable pre-warning system was built in this article,relevant machine vision detection algorithms were introduced,facial recognition and human dangerous action recognition technology and their application based on embedded platforms was studied.The main research work of this article includes following aspects:1)A wearable pre-warning platform overall structure based on machine vision was designed,with analyzed current facial recognition and human action recognition techniques.Kinect 2.0 was used to collect color and depth images of human dangerous action and skeletal point information.Sets of dangerous human action based on colored images,depth images,and skeletal point information data was constructed.Images of human face were harvested with camera to build a human face datasheet for the basement of following facial recognition pre-warning and dangerous action detecting system.2)Facial recognition technology was researched.The Adaboost algorithm based on Haar features is used to detect faces within the image collected by cameras.A convolutional neural network is constructed for facial recognition and warning.3)Human action recognition technology was analyzed,human dangerous action warning technology based on 3D convolutional neural network was researched,and continuous multiple frames input of color and depth images is used to identify dangerous actions.Aiming at the characteristics of Kinect human skeletal point capture,a pre-warning method based on LSTM for dangerous movements of human body using the feature of bone angle is proposed.4)Facial and human dangerous action pre-warning functions were achieved on Jeston TX2 embedded platform,relevant mechanical structure was designed according to the wearable requirements of the platform.The backpack-style container of the embedded platform was 3-D printed.Functions of the wearable pre-warning platform was verified through experimental verification results.