Similitude search

Similitude search allows you to find in the archive similar images using a certain track or an image.

Attention!

The processor must support the AVX2 or AVX512 instruction set (see ark.intel.com, Hardware requirements for neural analytics operation).
To use Similitude search, it is necessary to configure the Neural tracker by enabling the Similitude search option in it, and setting the time required for the algorithm to process the track in the Time of processing similitude track (sec) field.

You can search by images in the following way:

Set the search interval (see Setting a search interval).
Select the Similitude search type (see Selecting the search type).
Select a search method:
1. Search for a similar image by track:
  1. Select a track by left-clicking an object in the Surveillance window (1). When you use Similitude search, the image track is highlighted and its dimensions are displayed in the Surveillance window. The track image is displayed in the Similitude search type window (2).
  2. Enter the minimum level of similarity (in percent) of the selected track with images in the archive (3). The search results display only those videos that have images with a similarity level greater than the specified one. To remove a track from the search, click the button.
2. Search for a similar image by image:
  1. Select an image that is similar to the image you want to find in the archive (4). Supported formats: png, jpg, jpeg, jpe. If no image is selected, the search results display all images recognized within the specified time range. To clear the search criteria, click the button.
  2. Enter the minimum level of similarity (in percent) of the selected image with images in the archive (3). The search results display only those videos that have images with a similarity level greater than the specified one.
    Note
    When you search for a person, make sure that the track or image displays the person in full height. If the image is not complete, then incorrect recognitions can be included in the search results.
    If the search does not give required results at the selected level of similarity, then reduce its value. Note that lowering the similarity level below 50% can lead to incorrect recognitions.
Select the type of search results sorting: —by similarity level; —by time (5).
Click the Search button.
Note
Once you start the search, you can stop it at any time. To do this, click the Stop button, which is displayed in place of the Search button.

The search according to the specified criteria will start.

Note

The first time you search by images, the results appear with some delay (about three minutes).

Next time, results can be delayed for other reasons. For example, if there is a lot of data in the archive or if the hard disk is overloaded.

The search results panel displays video frames with images that match the search criteria. The recognized image is outlined with a red border, and the level of similarity is displayed below.

To quickly export an image from a frame:

Click on the image track in the Surveillance window.
Click the button (6).

The image is saved to the folder specified in the settings (see Configuring export options).

Use case for Similitude search

AI analytics extracts human appearance metadata and stores it in a database. Sorting through metadata sourced from multiple cameras, an intelligent algorithm identifies and presents similar individuals. Having said that, if a person is wearing a backpack, bag, or hat for example, the engine searches for people who have these same characteristics.

Page tree

Documentation for Axxon One 2.0. Documentation for other versions of Axxon One is available too.

Use case for Similitude search