Item monitoring for doorbell cameras

1 . A computer-implemented method comprising:

accessing a video stream comprising:

a pre-event image generated before a first activity, and

a post-event image generated after the first activity;

providing the pre-event image and the post-event image as an input to a machine-learning (ML) model trained for item detection based on statistical subtraction of image pixel data across a plurality of images that includes images captured before a training activity and after the training activity;

receiving, from the ML model, output that applies an image bounding box to a candidate item in at least a portion of the post-event image;

detecting that the candidate item was placed at a property using the output of the ML model that applies the image bounding box to the candidate item in at least the portion of the post-event image; and

determining that the candidate item is likely a package delivered to the property based on analysis of image pixel data of the candidate item within the image bounding box applied to at least the portion of the post-event image.

2 . The method of claim 1 , comprising:

computing differences in pixel values within the image pixel data, wherein the difference is computed between:

i) pixel values in the post-event image that are bounded by the image bounding box, and

ii) pixel values in the pre-event image.

3 . The method of claim 1 , comprising:

for a first time stamp, obtaining the pre-event image of an area of interest (AOI) having a boundary that overlaps with a threshold distance from an imaging device at the property; and

for a second time stamp, obtaining the post-event image of the AOI with respect to the boundary.

4 . The method of claim 3 , wherein:

the AOI includes a pre-event AOI that overlaps a portion of an area depicted in the pre-event image; and

the AOI coincides with a field of view of the imaging device used to generate an input video stream.

5 . The method of claim 4 , comprising:

generating, by obtaining a difference between pixel values in the pre-event and post-event images, a difference image;

generating, by removing pixels from the difference image, a cropped difference image;

generating, utilizing the cropped difference image, an image change measure score; and

utilizing, in the analysis of image pixel data, the image change measure score to determine that the candidate item is the package delivered to the property.

6 . The method of claim 5 , comprising:

in response to processing the post-event image using the ML model, detecting, from the post-event image, that the candidate item was placed at the property; and

in response to detecting that the candidate item was placed at the property, generating, using the ML model, the image bounding box as the overlay in the post-event image to outline the candidate item.

7 . The method of claim 1 , comprising:

computing a region-based similarity score indicating similarity between respective regions of the pre-event image and the post-event image.

8 . The method of claim 7 , comprising:

in response to determining the region-based similarity score, determining that a first item was delivered to the property or that a second item was removed after being delivered to the property.

9 . The method of claim 7 , wherein the region-based similarity score is computed from pixels within the bounding box.

10 . A system comprising:

a processing device; and

a non-transitory machine-readable storage device storing instructions that are executable by the processing device to cause performance of operations comprising:

accessing a video stream comprising:

a pre-event image generated before a first activity, and

a post-event image generated after detection of the first activity;

processing the pre-event and post-event image by providing the pre-event and post-event image as an input to a machine-learning (ML) model trained for package detection based on statistical subtraction of image pixel data across a plurality of images that include images captured before a training activity and after the training activity;

detecting that a candidate item was placed at a property using the output of the processing of the pre-event and the post-event image by the ML model;

in response to detecting that the candidate item was placed at the property, generating, using the ML model, an image bounding box that outlines the candidate item as an overlay in the post-event; and

determining that the candidate item is likely a package delivered to the property based on analysis of image pixel data of the candidate item within the image bounding box overlaid on the post-event.

11 . The system of claim 10 , comprising:

computing differences in pixel values within the image pixel data, wherein the difference is computed between:

i) pixel values in the post-event image that are bounded by the image bounding box, and

ii) pixel values in the pre-event image.

12 . The system of claim 10 , comprising:

for a first time stamp, obtaining the pre-event image of an area of interest (AOI) having a boundary that overlaps a threshold distance from an imaging device at the property; and

for a second time stamp, obtaining the post-event image of the AOI with respect to the boundary.

13 . The system of claim 12 , wherein:

the AOI includes a pre-event AOI that overlaps a portion of an area depicted in the pre-event image; and

the AOI coincides with a field of view of the imaging device used to generate an input video stream.

14 . The system of claim 13 , comprising:

generating, by obtaining a difference between pixel values in the pre-event and post-event images, a difference image;

generating, by removing pixels from the difference image, a cropped difference image;

generating, utilizing the cropped difference image, an image change measure score; and

utilizing, in the analysis of image pixel data, the image change measure score to determine that the candidate item is the package delivered to the property.

15 . The system of claim 14 , comprising:

in response to processing the post-event image using the ML model, detecting, from the post-event image, that the candidate item was placed at the property; and

in response to detecting that the candidate item was placed at the property, generating, using the ML model, the image bounding box as the overlay in the post-event image to outline the candidate item.

16 . The system of claim 10 , comprising:

computing a region-based similarity score indicating similarity between respective regions of the pre-event image and the post-event image.

17 . The system of claim 16 , comprising:

in response to determining the region-based similarity score, determining that a first item was delivered to the property or that a second item was removed after being delivered to the property.

18 . One or more non-transitory machine-readable storage devices storing instructions that are executable by one or more processing devices to cause performance of operations comprising:

accessing a video stream comprising:

a pre-event image generated before a first activity, and

a post-event image generated after detection of the first activity;

processing each of the pre-event and post-event image by providing the pre-event and post-event as an input to a machine-learning (ML) model trained for package detection based on statistical subtraction of image pixel data across a plurality of images that include images captured before a training activity and after the training activity;

detecting that a candidate item was placed at a property using the output of the processing of the pre-event and the post-event image by using the ML model;

in response to detecting that the candidate item was placed at the property, generating, using the ML model, an image bounding box that outlines the candidate item as an overlay in the post-event image; and

determining that the candidate item is likely a package delivered to the property based on analysis of image pixel data of the candidate item within the image bounding box overlaid on the post-event image.

19 . The one or more storage devices of claim 18 , wherein the operations comprise:

computing differences in pixel values within the image pixel data, wherein the difference is computed between:

i) pixel values in the post-event image that are bounded by the image bounding box, and

ii) pixel values in the pre-event image.

20 . The one or more storage devices of claim 19 , wherein the operations comprise:

for a first time stamp, obtaining the pre-event image of an area of interest (AOI) having a boundary that overlaps a threshold distance from an imaging device at the property; and

for a second time stamp, obtaining the post-event image of the AOI with respect to the boundary.