Search

CN-121982626-A - Lost article detection method based on monocular camera vision

CN121982626ACN 121982626 ACN121982626 ACN 121982626ACN-121982626-A

Abstract

The disclosure provides a lost article detection method based on monocular camera vision, which comprises the steps of acquiring a monitoring image of an area to be detected based on the monocular camera in real time, determining a first target area in the monitoring image according to depth information of the monitoring image, determining a second target area in the monitoring image according to texture characteristics of the monitoring image, determining a target area where a lost article is located according to the first target area and/or the second target area, and highlighting the target area in the monitoring image through a mark frame. According to the method and the device, the first target area and the second target area are respectively determined by combining the depth information difference and the local texture feature difference carried in the monitoring image acquired by the monocular camera, so that the rapid positioning and distinguishing between the first target area and the background area are realized, the target area where the lost article is positioned is finally determined, additional data labeling and sample training are not needed, the influence of ambient light is avoided, and rapid lost article identification can be realized.

Inventors

  • REN YI
  • SUN JIE
  • WANG FEI

Assignees

  • 中建材信息技术股份有限公司
  • 中建材信云智联科技有限公司
  • 中建材信息科技有限公司
  • 中建材信云智联科技有限公司北京分公司
  • 中建材信云智联科技(北京)有限公司

Dates

Publication Date
20260505
Application Date
20251219

Claims (10)

  1. 1. A lost article detection method based on monocular camera vision, comprising: Acquiring a monitoring image of an area to be detected in real time based on a monocular camera; Determining a first target area in the monitoring image according to the depth information of the monitoring image; determining a second target area in the monitoring image according to the texture characteristics of the monitoring image; and determining a target area where the lost article is located according to the first target area and/or the second target area, and highlighting the target area in the monitoring image through a mark frame.
  2. 2. The missing article detection method of claim 1, wherein the determining a first target area in the monitoring image based on depth information of the monitoring image includes: determining a depth value for each pixel in the monitor image; Extracting pixels, of which the difference value between the depth value and the preset depth value is larger than a first threshold value, as candidate pixels based on the preset depth value; and forming at least one communication area as the first target area according to the coordinates of all the candidate pixels in the monitoring image.
  3. 3. The missing article detection method of claim 2, wherein the determining a second target area in the monitoring image based on texture features of the monitoring image includes: Extracting texture features of the monitoring image based on a preset sliding window according to a preset step length, and determining local texture description features of all pixels in the window after each sliding; performing similarity calculation on each local texture description feature and the reference texture description feature of the corresponding window in the background texture feature library; taking a pixel area corresponding to the local texture description characteristic with the similarity smaller than a second threshold value as a suspected area; And integrating the suspected areas to form at least one second target area.
  4. 4. The missing item detection method of claim 3, wherein the local texture descriptive feature is extracted based on at least one of local binary pattern extraction, gray level co-occurrence matrix extraction, gradient histogram extraction.
  5. 5. The lost article detection method according to claim 3, wherein the determining the target area in which the lost article is located according to the first target area and/or the second target area comprises: Detecting whether the first target area and the second target area with the contact ratio larger than a third threshold exist according to the positions of the first target area and the second target area in the monitoring image; and when the first target area and the second target area with the overlap ratio larger than a third threshold value exist, taking a merging area of the first target area and the second target area as the target area.
  6. 6. The lost article detection method according to claim 3, wherein the determining the target area in which the lost article is located according to the first target area and/or the second target area comprises: and determining the second target area as the target area under the condition that the similarity between the local texture descriptive characteristic of the second target area and the reference texture descriptive characteristic is smaller than a fourth threshold value.
  7. 7. The lost article detection method according to any one of claims 1 to 6, further comprising: And pushing the monitoring image with the marking frame to a monitoring center and/or a field personnel operation terminal.
  8. 8. A lost article detection device based on monocular camera vision, comprising: the acquisition module is used for acquiring a monitoring image of the area to be detected by the monocular camera in real time; the depth information extraction module is used for determining a first target area in the monitoring image according to the depth information of the monitoring image; the texture feature extraction module is used for determining a second target area in the monitoring image according to the texture features of the monitoring image; and the target area marking module is used for determining the target area where the lost article is located according to the first target area and/or the second target area, and highlighting the target area in the monitoring image through a marking frame.
  9. 9. A storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of a lost article detection method based on monocular camera vision as claimed in any one of claims 1 to 7.
  10. 10. An electronic device comprising at least a memory, a processor, the memory having stored thereon a computer program, characterized in that the processor, when executing the computer program on the memory, implements the steps of a lost article detection method based on monocular camera vision as claimed in any one of claims 1 to 7.

Description

Lost article detection method based on monocular camera vision Technical Field The disclosure relates to the technical field of computer vision, in particular to a lost article detection method and device based on monocular camera vision, a storage medium and electronic equipment. Background With the rapid development of computer vision technology, the image processing-based article carry-over detection technology is widely applied in the fields of security monitoring, public safety and the like. The existing mainstream article carry-over detection method is mainly based on a two-dimensional image processing technology, a detection model is trained by collecting image data of a monitoring area, preprocessing and feature extraction are carried out on the image, and whether article carry-over exists or not is finally judged. However, since the types of the left-over articles are various, and the detection environment is complex and changeable, the traditional method is easy to generate false alarm and missing report under the conditions of light change, pedestrian shielding and the like. In addition, although the detection accuracy is improved by the detection method based on deep learning, a large amount of annotation data is needed for training, and when a new class of articles appears, the generalization capability of the model is limited, so that the practical application requirements are difficult to meet. Disclosure of Invention An object of an embodiment of the present disclosure is to provide a lost article detection method, apparatus, storage medium and electronic device based on monocular camera vision, which are used for solving the problems existing in the prior art. The lost article detection method based on monocular camera vision comprises the following steps of collecting a monitoring image of an area to be detected in real time based on the monocular camera, determining a first target area in the monitoring image according to depth information of the monitoring image, determining a second target area in the monitoring image according to texture features of the monitoring image, determining a target area where a lost article is located according to the first target area and/or the second target area, and highlighting the target area in the monitoring image through a marking frame. In some embodiments, the determining the first target area in the monitoring image according to the depth information of the monitoring image comprises determining a depth value of each pixel in the monitoring image, extracting pixels, of which the difference value between the depth value and the preset depth value is larger than a first threshold value, as candidate pixels based on the preset depth value, and forming at least one connected area as the first target area according to the coordinates of all the candidate pixels in the monitoring image. In some embodiments, the determining the second target area in the monitored image according to the texture features of the monitored image includes extracting the texture features of the monitored image based on a preset sliding window according to a preset step length, determining local texture description features of all pixels in the window after each sliding, performing similarity calculation on each local texture description feature and a reference texture description feature of a corresponding window in a background texture feature library, taking a pixel area corresponding to the local texture description feature with similarity smaller than a second threshold value as a suspected area, and integrating the suspected area to form at least one second target area. In some embodiments, the local texture descriptive feature is extracted based on at least one of local binary pattern extraction, gray level co-occurrence matrix extraction, gradient histogram extraction. In some embodiments, the determining the target area where the missing article is located according to the first target area and/or the second target area includes detecting whether the first target area and the second target area with the contact ratio greater than a third threshold exist according to positions of the first target area and the second target area in the monitored image, and taking a combined area of the first target area and the second target area as the target area when the first target area and the second target area with the contact ratio greater than the third threshold exist. In some embodiments, the determining the target area where the missing article is located according to the first target area and/or the second target area includes determining that the second target area is the target area if the similarity between the local texture descriptive feature of the second target area and the reference texture descriptive feature is less than a fourth threshold. In some embodiments, the method further comprises pushing the monitoring image with the marking frame to a monitoring center