A space occupancy person number counting method, a terminal device, and a storage medium

By acquiring panoramic images using a fisheye camera, a head and shoulder detection model is constructed and combined with foreground segmentation and moving target tracking. This solves the problems of false detection and missed detection in existing technologies and achieves highly accurate people counting in complex scenes.

CN116246298BActive Publication Date: 2026-05-29XIAMEN MILESIGHT IOT CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
XIAMEN MILESIGHT IOT CO LTD
Filing Date
2022-09-05
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing technologies suffer from false positives and false negatives in people counting, especially in complex scenarios. Pedestrian detection and head and shoulder detection methods based on target features and deep learning perform poorly in complex backgrounds or with high pedestrian traffic.

Method used

A fisheye camera is used to acquire panoramic images from above, and a head and shoulder detection model is constructed. Combined with foreground segmentation algorithm and moving target tracking, noise is eliminated by detecting branches of head and shoulder key points and statistical regularities. A fault tolerance mechanism is set up to reduce false detections and false negatives.

Benefits of technology

It improves the accuracy of headcount statistics, reduces the probability of false positives and false negatives, and ensures accurate headcount statistics in complex scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116246298B_ABST
    Figure CN116246298B_ABST
Patent Text Reader

Abstract

The application relates to a space occupancy number counting method, a terminal device and a storage medium, the method comprising the following steps: collecting overhead panoramic pictures to form a training set and marking human head shoulders; a head shoulder detection model is constructed, and the model is trained through the training set; each frame image in an overhead panoramic video is sequentially received, when a certain frame image is received, the head shoulder detection model after training is used for head shoulder detection on the frame image, and a foreground segmentation algorithm is used for foreground segmentation; the frame image and the image saved last time are subjected to motion target tracking, and a motion target queue is updated according to a motion target tracking judgment result; and a number counting result is obtained according to the head shoulder detection result and the motion target queue. On the basis of reserving a higher detection accuracy of a target detection network, dynamic and static information extracted by foreground segmentation is combined, the missing detection and false detection probability is greatly reduced, and the accuracy of number counting can be ensured.
Need to check novelty before this filing date? Find Prior Art