Research And Implementation Of Pedestrian Fall Detection For Complex Scenes

With the growth of the number of aging people in China,the tracking and detection of real-time behavioral trajectories of the elderly has become a hot spot in the current research of smart communities.This is because falls are one of the main causes of safety accidents among the elderly in their daily lives.According to statistics,about 30 million elderly people are injured by falls worldwide every year.And for those older adults who live in the community or independently,the risk of falls is even higher.In addition to the elderly,many people with disabilities who fall due to illness in low-traffic outdoor areas are also an important social safety hazard.Therefore,it is of great practical importance to study fall detection.Complex scenes generally refer to outdoor scenes in daily life,especially road intersections,playgrounds,parks and squares,and public activity facilities in the community,or isolated paths with many image environment levels,such as various alleyways in old neighborhoods.Unlike the indoor fall detection environment is relatively single,its interference term is relatively low,and the current algorithm is generally inefficient in detecting complex scenes.For example,cycling and shadows and other such closely related image data with human posture behavior can bring no small difficulties to the detection.The specific work of this paper is as follows:Firstly,starting from the visual algorithms in the two major research directions of fall detection,this article needs to logically decompose fall detection into motion recognition and posture recognition.A YOLOv5 roadside pedestrian fall detection method based on optical flow and posture is proposed to address the issue of low recognition accuracy caused by interference from electric vehicle flow and pedestrian shadows in existing pedestrian fall detection algorithms.Firstly,the original data is input into the YOLOv5 network through video frame extraction for preprocessing pedestrian video data monitoring,achieving pedestrian background reconstruction;Then extract reference frames for optical flow and pedestrian posture as their motion features;Finally,this feature is determined,and a fall detection network based on information fusion is used for fall feature recognition.Comparative experiments are conducted on different frame sequences and backgrounds.The method proposed in this article was tested on the pedestrian fall dataset Multiple Cameras Fall and Le2 i,and the results showed that the Precision and Recall of our method were significantly improved by about 10% higher than traditional methods in scenarios based on electric vehicle flow and pedestrian shadow interference.Then,after the above design and based on this,the pedestrian safety monitoring system is implemented.The back-end algorithm is based on Python,the development framework is Django,the database is MySQL,and the front-end is developed with Javascrip,Vue,and other related web technologies.The front-end and back-end use Nginx for data transit transmission,and the data is divided by algorithmic processing of the initial transmission data afterwards.The system realizes the main functions of simultaneous space and multi-lot monitoring,data visualization,abnormal data warning and unified data management for community streets,centralizing the display of community information for community managers,and providing timely decision-making.Integrating the above,this paper designs a more complete process for safe fall detection in complex scenes.