| With the increasingly serious problem of aging,nursing care for the elderly has become one of the current social research hotspots.As one of the most threatening factors to the health of the elderly,falling will not only seriously affect the physical and mental health of the elderly,but also bring enormous pressure to public health.Relying on manpower to care for the elderly has good effect,but it is inefficient and takes up a lot of medical resources.The existing visionbased fall detection systems generally use fixed cameras,which can not effectively monitor the target in its visual blind area.In view of the above problems,this paper studies the pedestrian tracking algorithm and the pedestrian instability attitude detection algorithm.First,the pedestrian tracking algorithm is studied.Taking consecutive multi-frame images as input,this paper mainly focuses on multi-target pedestrian tracking.Based on the DeepSORT algorithm,the target detection network is replaced,and the feature extraction network in the DeepSORT algorithm is improved by combining the lightweight module,so that the improved model has better tracking effect.Secondly,the pedestrian instability attitude detection algorithm is studied.Using video stream images as input,the pose of pedestrians is discriminated.This paper mainly judges whether pedestrians fall and other abnormal situations.Based on the YOLOv3-tiny network model,combined with the characteristics of the self-made data set,the backbone network is lightweight and designed to reduce the model parameters while ensuring the accuracy of the algorithm.And an improved attention mechanism module is added to the feature fusion layer,so that the improved network model has higher detection accuracy.Finally,this paper applies the algorithm to engineering practice,using the unmanned vehicle as the carrier,deploy the improved pedestrian tracking algorithm in this paper to the embedded device NVIDIA NANO on the unmanned vehicle,Combined with the vision scheme,the real-time location of the target is calculated,and the unmanned vehicle driving module is called to follow the pedestrian of a specific target,and the pedestrian instability attitude detection algorithm deployed on NVIDIA TX2 is used to detect pedestrians in real time.Experiments show that the algorithm can effectively detect pedestrian instability in practical applications.The experimental tests in different scenes and different lighting conditions prove that the pedestrian target tracking algorithm and pedestrian instability attitude detection algorithm designed in this paper have certain theoretical innovation and engineering practicability. |
