A People Counting Method Based on Trajectory Analysis

Abstract

The present invention is a people counting method based on trajectory analysis. The depth map of the scene is obtained by binocular camera or RGBD camera, the camera is calibrated, the depth map is converted into a three-dimensional point cloud by using the calibrated camera parameters, and the three-dimensional point cloud is Project to the X-Y plane to obtain the top view of the scene, use the head locking method proposed by the present invention to lock the head target in the top view, judge the trajectory by the detection surface, and judge the entry and exit of the trajectory. The invention is used for counting pedestrians in the scene (buses, escalators, passage openings, etc.), and accurately obtains the number of people entering and leaving.

Description

technical field [0001] The invention relates to a method for counting people based on trajectory analysis. Background technique [0002] People counting has always been a hot topic in monitoring systems, and there are many methods, such as the method of using laser rays, if the rays are blocked, someone will pass through, and there is also a method of using ultrasonic waves to measure the distance through ultrasonic waves to detect Pedestrians, however, these methods are not able to identify the direction of pedestrians, so, can not be widely used in people counting. There are many technical difficulties in using a monocular camera to count the number of pedestrians in a scene. In a simple scene, the number of people can be counted very well, but in a very complex scene, due to the influence of occlusion and false targets, it cannot be very accurate. Count the number of people in the scene. At the same time, there are no very stable pedestrian features to use under the mono...

AI Technical Summary

Problems solved by technology

[0002] People counting has always been a hot topic in monitoring systems, and there are many methods, such as the method of using laser rays, if the rays are blocked, someone will pass through, and there is also a method of using ultrasonic waves to measure the distance through ultrasonic waves to detect Pedestrians, however, these methods cannot identify the direction of pedestrians, so they cannot be widely used in people counting

There are many technical difficulties in using a monocular camera to count the number of pedestrians in a scene. In a simple scene, the number of people can be counted very well, but in a very complex scene, due to the influence of occlusion and false targets, it cannot be very accurate. Count the number of people in the scene. At the same time, there are no very stable pedestrian features to use under the monocular camera, and many reasons such as changes in light also cause the accuracy of the people counting to be low. Based on the defects of the monocular camera, you can use binocular or The RGB-D camera counts the number of people. In a crowded environment, pedestrians have very obvious characteristics, that is, the pedestrian's head must be higher than the shoulders. Based on this feature, it is easy to lock the head. After the matching stage, a trajectory is formed. These trajectories have a certain pattern, and the pattern of the trajectory is extracted by machine learning methods, so as to judge the behavior of pedestrians, and finally count the number of people in the passage

Images