Search

CN-122023760-A - Identifying potential false alarm detection frame

CN122023760ACN 122023760 ACN122023760 ACN 122023760ACN-122023760-A

Abstract

Identifying potential false positive detection boxes is disclosed. The present disclosure relates to an anchor-based object detection system and method that identifies potential false positive detections among three detection frames in the same image frame. Each detection box has a prediction IoU score that represents the confidence that it captures the object. The system determines (S502) an overlap of the second frame with the first frame and the third frame, wherein the second frame is located therebetween. If it is determined (S506) that the IoU score of the second box is below the set threshold, and if it is determined (S508) that the corresponding reference points within each box are substantially aligned, the reference point of the second box is close to the alignment line defined by the first box and the third box, it identifies (S504) the second box as a potential false positive.

Inventors

  • Emmanuel Hasselberg
  • Richard. Alebek
  • Yonatan Eriksson

Assignees

  • 安讯士有限公司

Dates

Publication Date
20260512
Application Date
20251028
Priority Date
20241111

Claims (15)

  1. 1. A method (500) for identifying potentially false positive detection boxes in a set of three detection boxes (202, 204, 206) within an anchor-based object detection system (402), wherein each of the three detection boxes is detected in a same image frame (200), wherein each of the three detection boxes is associated with a respective prediction intersection ratio IoU score, wherein the prediction IoU score indicates a confidence that a detection box represents an object of the anchor-based object detection system, the method comprising: determining (S502) that a second detection frame (206) at least partially overlaps both a first detection frame (202) and a third detection frame (204), wherein the second detection frame is located between the first detection frame and the third detection frame in the image frame; Identifying (S504) that the second detection box is a potential false positive detection box by: determining (S506) that the predicted IoU score for the second detection box is below a first threshold score, and -Determining (S508) a first reference point (208) in the first detection frame, a corresponding second reference point (210) in the second detection frame and a corresponding third reference point (212) in the third detection frame, and-determining that the first reference point, the second reference point and the third reference point are substantially aligned in the image frame such that the second reference point is within a threshold distance from an alignment defined by the first reference point and the third reference point.
  2. 2. The method of claim 1, wherein determining that the first reference point, the second reference point, and the third reference point are substantially aligned comprises: Determining a first vector (214) between a first pair of reference points selected from the first reference point, the second reference point and the third reference point and a second vector (216) between a second, different pair of reference points selected from the first reference point, the second reference point and the third reference point, wherein the first reference point, the second reference point and the third reference point are substantially aligned in the image frame if an absolute value of a cosine of an angle (θ) between the first vector and the second vector is less than a threshold from 1.
  3. 3. The method of claim 1, wherein determining that the first reference point, the second reference point, and the third reference point are substantially aligned comprises: determining that the second reference point is within a threshold distance from a line formed between the first reference point and the third reference point.
  4. 4. The method of claim 1, wherein identifying that the second detection box is a potential false positive detection is further performed by: Determining (S510) that the predicted IoU score for the second detection box is at least a second threshold score lower than the predicted IoU score for each of the first detection box and the third detection box.
  5. 5. The method of claim 1, wherein each of the three detection boxes is associated with a predicted object class, wherein identifying the second detection box as a potential false positive detection is further performed by: determining (S512) that the prediction object class associated with each of the first, second, and third detection boxes is the same.
  6. 6. The method of claim 1, wherein the first reference point, the second reference point, and the third reference point are midpoints of top edges of the first, second, and third detection frames, respectively.
  7. 7. The method of claim 1, wherein the first reference point, the second reference point, and the third reference point are center points of the first detection frame, the second detection frame, and the third detection frame, respectively.
  8. 8. The method of claim 1, further comprising: A lower probability associated with an object trajectory in an object tracking system (406) is assigned to the second detection box than to the first detection box and the third detection box.
  9. 9. The method of claim 8, wherein assigning a lower probability comprises assigning a higher cost associated with the object trajectory for the second detection box than for the first detection box or the third detection box.
  10. 10. The method of any of claims 8 and 9, wherein assigning a lower probability comprises assigning the second detection box to a lower priority partition of detection boxes associated with the object trajectory and assigning the first detection box and the third detection box to a higher priority partition of detection boxes associated with the object trajectory, wherein each partition is sequentially processed to be associated with a trajectory in the object tracking system.
  11. 11. The method of claim 1, further comprising: The second detection box marked as a potential new object trajectory in the object tracking system is filtered out of the set of detection boxes in the first image frame.
  12. 12. The method of claim 1, further comprising: The first detection box and the third detection box are counted as confirmed objects in an object counting system (408), and the second detection box is counted as an uncertain object in the object counting system.
  13. 13. The method of claim 1, further comprising: the second detection box is filtered out of an initial set of detection boxes including the first detection box, the second detection box, and the third detection box, and a remaining set of detection boxes is used in a downstream analysis system (404).
  14. 14. A non-transitory computer readable storage medium having stored thereon instructions for implementing the method according to any of claims 1 to 13 when executed on one or more devices having processing capabilities.
  15. 15. An anchor-based object detection system (402) configured to identify potential false positive detection boxes in a set of three detection boxes (202, 204, 206), wherein each of the three detection boxes is detected in a same image frame (200), wherein each of the three detection boxes is associated with a respective prediction intersection ratio IoU score, wherein the prediction IoU score indicates a confidence of the anchor-based object detection system that the detection box represents an object, the anchor-based object detection system configured to: determining (S502) that a second detection frame (206) at least partially overlaps both a first detection frame (202) and a third detection frame (204), wherein the second detection frame is located between the first detection frame and the third detection frame in the image frame; Identifying (S504) that the second detection box is a potential false positive detection box by: Determining (S506) that the IoU score of the second detection box is below a first threshold score, and -Determining (S508) a first reference point (208) in the first detection frame, a corresponding second reference point (210) in the second detection frame and a corresponding third reference point (212) in the third detection frame, and-determining that the first reference point, the second reference point and the third reference point are substantially aligned in the image frame such that the second reference point is within a threshold distance from an alignment defined by the first reference point and the third reference point.

Description

Identifying potential false alarm detection frame Technical Field The present disclosure relates to object detection, and in particular, to methods, systems, and software for identifying potentially false positive detection boxes in an anchor-based object detection system. Background In modern object detection systems such as single-shot detectors (SSD) and YOLO (you only see once), the anchor boxes are the basis for detecting objects on the image. These anchor boxes are predefined and typically overlay the image in various proportions and aspect ratios to detect objects of different sizes and shapes. During training, the object detection system learns to adjust these anchor boxes to better accommodate objects by encoding those objects that have a high intersection ratio (IoU) score, ioU score representing the overlap between the anchor boxes and the real data object. For training purposes, anchor boxes with significant IoU overlaps are assigned to the object. Significant problems can occur when multiple objects are positioned close to each other, or when anchor boxes are sparsely distributed over the image. In this case, more than one object may have similar IoU with a particular anchor box, resulting in ambiguous assignments during training. Such ambiguity may lead to a phenomenon known as intermediate detection. When two or more objects share a similar IoU score with the same anchor box, the object detection system may assign the anchor boxes to different objects non-uniformly during training. This results in intermediate detection that is false positive detection or ambiguous detection (false detection box between real objects). These intermediate detections negatively impact the performance of the object detection system by introducing false positives. Thus, there is a need for improvement in this context. Disclosure of Invention In view of the above, it would be advantageous to solve or at least mitigate one or more of the above-mentioned disadvantages, as set out in the accompanying independent patent claims. According to a first aspect of the present invention there is provided a method for identifying potentially false positive detection boxes in a set of three detection boxes within an anchor-based object detection system, wherein each of the three detection boxes is detected in the same image frame, wherein each of the three detection boxes is associated with a respective prediction cross ratio (IoU) score, wherein the prediction IoU score indicates a confidence that the detection boxes represent an object of the anchor-based object detection system, the method comprising determining that a second detection box at least partially overlaps both the first detection box and a third detection box, wherein the second detection box is located between the first detection box and the third detection box in the image frame, identifying that the second detection box is potentially false positive detection box by determining that the prediction IoU score of the second detection box is below a first threshold score, and determining a first reference point in the first detection box, a corresponding second reference point in the second detection box, and a corresponding third reference point in the third detection box, and determining that the first reference point, the second reference point, and the third reference point are at a distance from the first reference point and the third reference point are substantially aligned within a threshold. The present disclosure addresses the problem of intermediate detection caused by ambiguous anchor frame assignments during training of an anchor-based object detection system, particularly where anchor frames are sparsely populated. The techniques described herein aim to improve detection accuracy by identifying potentially false positive detection boxes caused by such ambiguity. In particular, the method focuses on identifying detection boxes that may fall between real objects. The method introduces a strategy for managing ambiguous detection boxes while minimizing computational impact, thereby allowing the object detection system to maintain reliable performance even under hardware constraints or when anchor boxes are sparsely distributed. By identifying these ambiguous detection boxes, the object detection system is better equipped to handle false positives, thereby improving accuracy and efficiency. The "prediction IoU score" in an object detection system like SSD or YOLO refers to a measure of the prediction by the object detection system/model that indicates the extent to which the proposed detection (bounding box) may overlap with the actual object in the image. IoU (cross-over) conventionally refers to the ratio of the area of overlap between the prediction bounding box and the real data box divided by the area of their union. However, in this scenario, the prediction IoU score serves as a confidence measure that predicts the likelihood that the bounding box g