Methods and devices for setting a threshold in an object detection system

1 . A method for setting a threshold in an object detection system, the method comprising:

a) detecting objects in a plurality of images depicting a scene and storing outputs, wherein the stored outputs comprise one or more object detections of objects classified as a first object class, each object detection being associated with an object location and a confidence score representing a confidence that the object detection is of the first object class;

b) for each region of a plurality of regions in the scene:

based on the stored outputs, determining a representative confidence score of object detections classified as the first object class and being associated with an object location within said region, wherein the representative confidence score is an average, or a weighted average, of the confidence scores of a plurality of object detections being associated with an object location within said region in the stored output;

c) operating on the stored outputs to remove any object detection of the one or more object detections being associated with a confidence score below a first threshold, wherein the method further comprises:

d) setting the first threshold for object detections associated with an object location within a first region of the plurality of regions based on the representative confidence score determined for the first region, wherein the first threshold is set to have a lower value if a higher representative confidence score has been determined for the first region compared to if a lower representative confidence score has been determined for the first region,

wherein

setting the first threshold comprises determining a difference between the representative confidence score and a predefined threshold confidence score, wherein upon the difference being above 0, the first threshold is set to a value lower than the predefined threshold confidence score, wherein upon the difference being below 0, the first threshold is set to a value higher than the predefined threshold confidence score, wherein the predefined threshold confidence score is a baseline confidence score.

2 . The method according to claim 1 , wherein determining a representative confidence score comprises disregarding any object detections associated with a confidence score below a second threshold when determining the representative confidence score.

3 . The method according to claim 1 , wherein a deviation between the first threshold and the predefined threshold confidence score increases with increasing size of the difference.

4 . The method according to claim 3 , wherein the value of the first threshold is capped at a minimum value and a maximum value.

5 . The method according to claim 1 , wherein, for each region of the plurality of regions, the representative confidence score is an average confidence score of object detections being associated with an object location within said region.

6 . The method according to claim 1 , wherein step a) further comprises associating each of the one or more object detections with a respective timestamp, wherein, for each region of the plurality of regions, the representative confidence score is a weighted average confidence score of object detections being associated with an object location within said region, wherein a weight is based on one or more of: a value of the confidence score, and the associated timestamp.

7 . The method according to claim 1 , wherein step a) further comprises associating each of the one or more object detections with a respective timestamp, wherein the method further comprises:

e) for each region of the plurality of regions in the scene:

determining a latest timestamp for the region, by determining a latest timestamp in the stored output of an object detection being associated with an object location within said region; and

f) adjusting the first threshold set in step d) for object detections associated with an object location within the first region of the plurality of regions based on the latest timestamp determined for the first region, wherein the first threshold is adjusted to have a higher value if an earlier latest time stamp has been determined for the first region, compared to if a later latest time stamp has been determined for the first region.

8 . The method according to claim 1 , further comprising the step of:

continuously receiving images;

wherein the steps a)-d) are performed at a plurality of points in time, wherein for each point in time, step a) of detecting objects is performed on the latest received N>1 images.

9 . The method of claim 7 , further comprising the step of:

continuously receiving images;

wherein the steps a)-f) are performed at a plurality of points in time, wherein for each point in time, step a) of detecting objects is performed on the latest received N>1 images.

10 . The method of claim 1 , wherein step d) is performed for M>1 regions of the plurality of regions in the scene, the method further comprising the step of:

receiving, a first object detection classified as the first object type, the object detection defined by a bounding box and being associated with and a confidence score, wherein said bounding box overlaps the M regions,

determining the first threshold to be used for said first object detection based on the first threshold set for each region of the M regions.

11 . The method of claim 10 , wherein the step of determining the first threshold to be used for said first object detection comprises one of:

calculating an average value of the first thresholds set for each of the M regions and using the average value as the first threshold for the first object detection;

calculating a maximum value of the first thresholds set for each of the M regions and using the maximum value as the first threshold for the first object detection;

calculating a minimum value of the first thresholds set for each of the M regions and using the minimum value as the first threshold for the first object detection.

12 . An object detection system comprising:

circuitry configured for:

a) detecting objects in a plurality of images depicting a scene and storing outputs, wherein the stored outputs comprise one or more object detections of objects classified as a first object class, each object detection being associated with an object location and a confidence score representing a confidence that the object detection is of the first object class;

b) for each region of a plurality of regions in the scene:

based on the stored outputs, determining a representative confidence score of object detections classified as the first object class and being associated with an object location within said region, wherein the representative confidence score is an average, or a weighted average, of the confidence scores of a plurality of object detections being associated with an object location within said region in the stored output;

c) operating on the stored outputs to remove any object detection of the one or more object detections being associated with a confidence score below a first threshold, wherein the method further comprises:

d) setting the first threshold for object detections associated with an object location within a first region of the plurality of regions based on the representative confidence score determined for the first region, wherein the first threshold is set to have a lower value if a higher representative confidence score has been determined for the first region compared to if a lower representative confidence score has been determined for the first region,

wherein setting the first threshold comprises determining a difference between the representative confidence score and a predefined threshold confidence score, wherein upon the difference being above 0, the first threshold is set to a value lower than the predefined threshold confidence score, wherein upon the difference being below 0, the first threshold is set to a value higher than the predefined threshold confidence score, wherein the predefined threshold confidence score is a baseline confidence score.

13 . The object detection system of claim 12 , being implemented in a monitoring camera.

14 . A non-transitory computer-readable storage medium having stored thereon instructions for implementing a method, when executed on a device having processing capabilities, the method for setting a threshold in an object detection system, the method comprising:

a) detecting objects in a plurality of images depicting a scene and storing outputs, wherein the stored outputs comprise one or more object detections of objects classified as a first object class, each object detection being associated with an object location and a confidence score representing a confidence that the object detection is of the first object class;

b) for each region of a plurality of regions in the scene:

based on the stored outputs, determining a representative confidence score of object detections classified as the first object class and being associated with an object location within said region, wherein the representative confidence score is an average, or a weighted average, of the confidence scores of a plurality of object detections being associated with an object location within said region in the stored output;

c) operating on the stored outputs to remove any object detection of the one or more object detections being associated with a confidence score below a first threshold, wherein the method further comprises:

d) setting the first threshold for object detections associated with an object location within a first region of the plurality of regions based on the representative confidence score determined for the first region, wherein the first threshold is set to have a lower value if a higher representative confidence score has been determined for the first region compared to if a lower representative confidence score has been determined for the first region,

wherein setting the first threshold comprises determining a difference between the representative confidence score and a predefined threshold confidence score, wherein upon the difference being above 0, the first threshold is set to a value lower than the predefined threshold confidence score, wherein upon the difference being below 0, the first threshold is set to a value higher than the predefined threshold confidence score, wherein the predefined threshold confidence score is a baseline confidence score.