Fall pre-judgment method based on deep neural network and panoramic segmentation

Abstract

The invention provides a fall pre-judgment method based on the combination of a deep neural network and a panoramic segmentation method, the fall detection pre-judgment function can be efficiently and quickly achieved, the deep neural network and the image panoramic segmentation method are combined, short-term and real-time evaluation and notification are carried out on an imminent fall risk, and long-term behavior learning and prediction are carried out on future risks. According to the invention, a deep neural network (DNN) in deep learning is adopted to construct a panoramic segmentation network, and pixel-level segmentation is carried out on a video image in fall detection through an image panoramic segmentation algorithm, so that scene understanding between a caregiver and an environment where the caregiver is located is achieved, and fall pre-judgment on a dangerous environment is achieved.

Description

technical field [0001] The invention relates to the field of intelligent communication, in particular to a fall prediction method based on the combination of a deep neural network and an image panorama segmentation algorithm. Background technique [0002] At present, there are many discussions on the fall prediction method based on computer vision at home and abroad. According to the different algorithms and implementation methods, they can be divided into three categories: (1) Posture estimation: this method combines deep learning with recurrent neural network. Combined methods collect data on various postures of the human body to build a personal posture library, and use a series of posture actions to realize predictive alarms for possible falls. This method can achieve fall prediction to a certain extent, but it has a huge amount of calculation, high requirements on hardware equipment, it is difficult to achieve real-time detection, and it lacks recognition and understand...

AI Technical Summary

Problems solved by technology

This method realizes the scene understanding between the human body and the surrounding environment, and can effectively realize the fall prediction function, but the training method is relatively cumbersome, lacking a fast and accurate image segmentation algorithm, it is difficult to accurately extract the human body and the surrounding environment

In addition, using a deep neural network to train a large amount of image data will result in too much calculation and high energy consumption, making it difficult to achieve real-time prediction

[0003] Combining the current research status in the world analyzed above, it can be found that the current fall prediction method is facing the following problems: (1) The calculation required for the operation of the algorithm is relatively large, resulting in a low operating speed and unable to achieve real-time ; (2) Lack of fast and accurate image segmentation algorithm

Images