Parking position identification method based on infrared induction camera

Abstract

The invention relates to the field of computer vision, in particular to a staying position recognition method based on an infrared induction camera, and the method comprises the steps: collecting a moving image of a person in a target space through the infrared induction camera; using a deep learning algorithm to extract image coordinates of people in a range corresponding to the target space in each image; according to the image coordinate of the person and the image shooting time information, identifying the image coordinate of the staying position; and according to the projection coordinate range of the target space, converting the image coordinates of the stay position in the image into projection coordinates. According to the embodiment of the invention, the staying position of the crowd in the space can be flexibly and efficiently identified by designing the staying position identification algorithm based on the image shot by the infrared induction camera, the reconnaissance level of activity characteristics of the crowd in the scattered space is improved, light weight, low cost and accuracy of staying position identification are realized, and the intelligent level is higher.

Description

Technical Field

[0001] The present application relates to the field of computer vision, and in particular to a method for identifying a stop position based on an infrared sensing camera. Background Art

[0002] Staying behavior is a key factor in measuring the efficiency of using green space and public open space. In related technologies, image-based recognition of stay locations is usually based on videos taken by camera equipment. However, the deployment of related equipment involves problems such as power supply, data storage and transmission difficulties, which makes it difficult to popularize in a large number of public open spaces. There is a lack of lightweight, flexible stay location recognition methods with low data volume requirements, which needs to be solved urgently. Summary of the invention

[0003] The present application provides a method for identifying a stop position based on an infrared sensing camera to solve the problem of lack of a flexible stop position identification method that is lightweight and has low data volume requirements.

[0004] The present application provides a method for identifying a stop position based on an infrared sensing camera, comprising the following steps: using an infrared sensing camera to collect activity images of people in a target space; using a deep learning algorithm to extract image coordinates of people within a corresponding range of the target space in each image; identifying the image coordinates of the stop position based on the image coordinates of the person and image shooting time information; and converting the image coordinates of the stop position in the image into projection coordinates based on the projection coordinate range of the target space.

[0005] Optionally, in one embodiment of the present application, the use of an infrared sensing camera to capture images of human activities in a target space includes: confirming the scope of the target space to be analyzed, deploying an infrared sensing camera, and photographing the entire target space at a top-down angle of 30 to 90 degrees; using the infrared sensing camera within a certain time range, through an object infrared sensing mode, to photograph the target space with a minimum continuous shooting interval of 15 seconds to capture images of human activities.

[0006] Optionally, in one embodiment of the present application, the use of a deep learning algorithm to extract image coordinates of people within the corresponding range of the target space in each image includes: establishing a Cartesian coordinate system with the lower left corner of the image as the origin and pixels as units; identifying the minimum envelope rectangle of the corresponding range of the target space in the image, and recording the image coordinates of the four vertices; using the yolo-v8 deep learning algorithm to perform target detection on the image with people as the target, and recording the image coordinates of the four vertices of the target identification box of each person and the time when the photo was taken.

[0007] Optionally, in one embodiment of the present application, the image coordinates of the stopping position are identified based on the image coordinates of the person and the image shooting time information, including: grouping the image coordinates of the four vertices of the person's target recognition frame according to the image shooting recognition, and taking the value obtained by rounding down after dividing the shooting time in the Unix timestamp format by 15 seconds as the group number; comparing the rectangular ranges formed by the four vertices of the target recognition frames in adjacent groups with group numbers, and taking the target recognition frames with an overlap of more than 80% as the target recognition frames of the stopping pedestrian; calculating and recording the image coordinates of the midpoint of the lower edge of the target recognition frame of the stopping pedestrian, and using it as the image coordinates of the stopping position.

[0008] Optionally, in one embodiment of the present application, the converting the image coordinates of the stay position in the image into projection coordinates according to the projection coordinate range of the target space includes: establishing a minimum envelope rectangle according to the target space range, and dividing it into rectangular analysis units of equal size; establishing a Cartesian coordinate system with the southwest corner as the origin and meters as the unit according to the minimum envelope rectangle, and recording the projection coordinates of the four vertices; matching the image coordinates of the four vertices of the minimum envelope rectangle of the corresponding range of the target space in the image with the projection coordinates of the minimum envelope rectangle of the target space to obtain the image coordinates of each corresponding analysis unit in the image; counting the number of stay positions in each corresponding analysis unit in the image; and converting the number of stay positions in each corresponding analysis unit in the image into the number of stay positions in each analysis unit in the projection coordinates. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 A flowchart of a method for identifying a stop position based on an infrared sensing camera provided in an embodiment of the present application; Figure 2 A schematic diagram of using an infrared sensing camera to capture images of human activity in a target space according to an embodiment of the present application; Figure 3 This is a flowchart of using a deep learning algorithm to extract image coordinates of people within a corresponding range of a target space in each image according to an embodiment of the present application; Figure 4 This is a flow chart of identifying the image coordinates of a stay position according to the image coordinates of a person and the image shooting time information according to one embodiment of the present application; Figure 5 A flowchart of converting the image coordinates of the stop position in the image into the projection coordinates according to the projection coordinate range of the target space according to one embodiment of the present application; Figure 6 A schematic diagram of a method for obtaining image coordinates of each analysis unit in an image according to an embodiment of the present application. DETAILED DESCRIPTION

[0010] Figure 1 The overall process of this application is shown, including four main steps: image acquisition, target detection, dwell position recognition and coordinate conversion. The specific workflow is: Use infrared sensing cameras to collect images of people's activities in the target space; Use deep learning algorithms to extract the image coordinates of people within the corresponding range of the target space in each image; According to the image coordinates of the person and the image shooting time information, the image coordinates of the staying position are identified; According to the projection coordinate range of the target space, the image coordinates of the dwell position in the image are converted into projection coordinates.

[0011] Figure 2 This application demonstrates how to use an infrared sensor camera to collect images of people's activities in the target space. The specific workflow is: (1) Confirm the target space range to be analyzed, deploy infrared sensing cameras, and photograph the entire target space at a 30 to 90 degree downward angle; (2) Use an infrared sensing camera within a certain time range to capture images of human activities in the target space through the object infrared sensing mode with a minimum continuous shooting interval of 15 seconds.

[0012] Figure 3 This demonstrates how this application uses deep learning algorithms to extract the image coordinates of people in the corresponding range of the target space in each image. The specific workflow is: (3) Establish a Cartesian coordinate system with the lower left corner of the image as the origin and pixels as the unit; (4) Identify the minimum enveloping rectangle of the target space in the image and record the image coordinates of the four vertices; (5) Use the yolo-v8 deep learning algorithm to perform target detection on images with people as the target, and record the image coordinates of the four vertices of each person’s target recognition box and the time when the photo was taken.

[0013] Figure 4 This demonstrates how this application can identify the image coordinates of the location where a person is staying based on the image coordinates of the person and the image shooting time information. The specific workflow is as follows: (6) The image coordinates of the four vertices of the human target recognition frame are grouped according to the image shooting recognition, and the shooting time in Unix timestamp format is divided by 15 seconds and rounded down as the group number; (7) Compare the rectangular ranges formed by the four vertices of the target recognition frames in adjacent groups, and take the target recognition frame with an overlap of more than 80% as the target recognition frame of the stopped pedestrian; (8) Calculate and record the image coordinates of the midpoint of the lower edge of the target recognition box of the pedestrian who is staying, and use them as the image coordinates of the stopping position.

[0014] Figure 5 This demonstrates how this application converts the image coordinates of the stop position in the image into projection coordinates according to the projection coordinate range of the target space. The specific workflow is: (9) According to the target space range, establish the minimum enveloping rectangle and divide it into rectangular analysis units of equal size; (10) Based on the minimum enveloping rectangle, establish a Cartesian coordinate system with the southwest corner as the origin and meters as the unit, and record the projection coordinates of the four vertices; (11) Match the image coordinates of the four vertices of the minimum envelope rectangle of the target space corresponding to the image with the projection coordinates of the minimum envelope rectangle of the target space to obtain the image coordinates of each corresponding analysis unit in the image; (12) Counting the number of dwell positions in each corresponding analysis unit in the image; (13) Convert the number of dwell positions in each corresponding analysis unit in the image into the number of dwell positions in each analysis unit in the projection coordinates.

[0015] Figure 6 It shows how to obtain the image coordinates of each analysis unit in the image according to the coordinate system conversion algorithm. The specific workflow is as follows: (11-1) Matching the image coordinates of the four vertices of the minimum envelope rectangle of the target space corresponding to the image with the projection coordinates of the minimum envelope rectangle of the target space; (11-2) According to the vertex positions of the quadrilateral ABCD in the study space, extend AC, BD and BA, DC to intersect at the perspective vanishing points P and O respectively, and obtain the angle between OP and OD and the angle between OD and OB; (11-3) Extend OC into a ray and rotate Ω with O as the endpoint in the direction away from the vanishing point to obtain the lower edge of the analysis unit closest to the vanishing point among the n analysis units with P as the vanishing point in the image coordinates, and its intersection points E and F with AC and BD; (11-4) Consider the rest of the space as the lower edge of (n-1) analysis units centered at the vanishing point P, and obtain the lower edge line of the second analysis unit and its intersection points G and H with AC and BD by the method in (12-2); repeat the above method to obtain the lower edge lines of all n analysis units centered at the vanishing point P and all their intersection points with AC and BD; (11-5) By the method in (11-3), obtain the lower edge of the analysis unit closest to the vanishing point among the m analysis units with point O as the vanishing point in the image coordinates, and its intersection points I and J with CD and AB; repeat the above method to obtain the lower edge lines of all m analysis units centered on the vanishing point O, and all their intersection points with CD and AB; (11-6) Determine the image coordinates of all vertices of the analysis units according to the upper and lower edge lines of each analysis unit.

[0016] The formula for calculating the tangent value of Ω in (11-3) is:

Claims

1. A method for identifying a stop position based on an infrared sensing camera, characterized in that It can flexibly and efficiently identify the location of people staying in the space, including the following steps: Use infrared sensing cameras to collect images of people's activities in the target space; Use deep learning algorithms to extract the image coordinates of people within the corresponding range of the target space in each image; According to the image coordinates of the person and the image shooting time information, the image coordinates of the staying position are identified; According to the projection coordinate range of the target space, the image coordinates of the dwell position in the image are converted into projection coordinates.

2. The method according to claim 1, characterized in that The method of using an infrared sensing camera to collect activity images of people in the target space includes: (1) Confirm the target space range to be analyzed, deploy infrared sensing cameras, and photograph the entire target space at a 30 to 90 degree downward angle; (2) Using an infrared sensing camera within a certain time range, in the object infrared sensing mode, to shoot the target space with a minimum continuous shooting interval of 15 seconds to collect images of human activities; 3. The method according to claim 1, characterized in that The method of using a deep learning algorithm to extract the image coordinates of a person within the corresponding range of the target space in each image includes: (3) Establish a Cartesian coordinate system with the lower left corner of the image as the origin and pixels as the unit; (4) Identify the minimum enveloping rectangle of the target space in the image and record the image coordinates of the four vertices; (5) Use the yolo-v8 deep learning algorithm to perform human-targeted target detection on the image, and record the image coordinates of the four vertices of each person's target recognition box and the time when the photo was taken; 4. The method according to claim 1, characterized in that: The step of identifying the image coordinates of the stop position according to the image coordinates of the person and the image shooting time information includes: (6) The image coordinates of the four vertices of the human target recognition frame are grouped according to the image shooting recognition, and the shooting time in Unix timestamp format is divided by 15 seconds and rounded down as the group number; (7) Compare the rectangular ranges formed by the four vertices of the target recognition frames in adjacent groups, and take the target recognition frame with an overlap of more than 80% as the target recognition frame of the stopped pedestrian; (8) Calculate and record the image coordinates of the midpoint of the lower edge of the target recognition box of the pedestrian who is staying, and use them as the image coordinates of the stopping position; 5. The method according to claim 1, characterized in that The step of converting the image coordinates of the stop position in the image into projection coordinates according to the projection coordinate range of the target space includes: (9) According to the target space range, establish the minimum enveloping rectangle and divide it into rectangular analysis units of equal size; (10) Based on the minimum enveloping rectangle, establish a Cartesian coordinate system with the southwest corner as the origin and meters as the unit, and record the projection coordinates of the four vertices; (11) Match the image coordinates of the four vertices of the minimum envelope rectangle of the target space corresponding to the image with the projection coordinates of the minimum envelope rectangle of the target space to obtain the image coordinates of each corresponding analysis unit in the image; (12) Counting the number of dwell positions in each corresponding analysis unit in the image; (13) Convert the number of dwell positions in each corresponding analysis unit in the image into the number of dwell positions in each analysis unit in the projection coordinates.

Images