| CPC G06Q 10/063 (2013.01) [G06F 16/25 (2019.01); G06N 3/04 (2013.01); G06Q 50/26 (2013.01)] | 7 Claims |
|
1. A method for managing a people flow of a public place in a smart city, which is executed by a processor of at least one public place management platform, comprising:
obtaining pedestrian distribution information in a preset area during a current time period via network from a storage device;
determining, by processing the pedestrian distribution information through an area location prediction model, at least one area location in the preset area for a future time period, a population flow load of the area location being greater than a first threshold;
wherein the area location prediction model includes a graph neural network model;
a graph input into the graph neural network model includes at least two nodes and at least one edge, an edge characteristic of the at least one edge includes a relationship strength vector;
the relationship strength vector reflects a relationship strength between two nodes;
the relationship strength vector is determined from a plurality of vectors, wherein the plurality of vectors include distance strength vectors, location strength vectors, and category strength vectors;
the area location prediction model is obtained by training based on training data, the training data includes at least one second training sample and at least one second label, wherein each second training sample includes sample pedestrian distribution information, and each second label is a sample area location that is determined by an actual population flow load of the preset area in the sample pedestrian distribution information; and
the sample area location is an area location that the actual population flow load of the preset area in the sample pedestrian distribution information is greater than a first threshold;
generating, based on the area location, prompt information; and
feedbacking the prompt information to a user terminal of a user platform through a service platform via the network;
wherein the obtaining pedestrian distribution information in a preset area during a current time period via network from a storage device includes:
processing a plurality of videos based on a recognition model through a sensor network sub-platform; determining population flow information of a plurality of preset areas in the plurality of videos,
obtaining the pedestrian distribution information in the preset area during the current time period based on the population flow information and transmitting the pedestrian distribution information to the storage device through the sensor network sub-platform;
wherein the recognition model includes a Yolo model that recognizes users in the plurality of videos, the recognition model is obtained through a training process, and the training process comprising:
generating a plurality of first training samples and first labels; wherein the plurality of first training samples include sample videos obtained based on historical surveillance videos, the first labels include a sample object of the users and a category corresponding to a sample object box;
inputting the plurality of first training samples with the first labels into an initial recognition model;
updating parameters of the initial recognition model through training; and
obtaining the recognition model when the initial recognition model satisfies a preset condition;
wherein an output of the recognition model includes an image segmentation result, and the image segmentation result includes object boxes and categories corresponding to the object boxes; the method further includes:
determining whether a same user exists in a plurality of object boxes of a group of object boxes based on an object determination model; wherein the object determination model includes a convolutional neural network model and a deep neural network model, the convolutional neural network model processes the plurality of object boxes and outputs image characteristics corresponding to the plurality of object boxes; and the deep neural network model processes any two of the image characteristics corresponding to the plurality of object boxes and determines whether the object boxes corresponding to the two image characteristics are the object boxes corresponding to the same user;
in response to a determination that the same user exists in the plurality of object boxes, merging the object boxes corresponding to the same user and determining a count of merged object boxes in the group of object boxes; and
determining a count of users in the videos based on the count of merged object boxes.
|