Detection Filtering

The Detection Filtering feature makes it possible to filter the output of Head Detection, based on a number of criteria.

Detection Filtering can be performed at different stages of the analysis. The criteria of filtering can also be different at the different stages. UVAP supports the filtering at the following stages and by the criteria listed for each stage:

Detection filtering in Multi-Graph Runner:

The main purpose of this filtering is optimization: it helps optimize the quality or performance of other core detectors in Multi-Graph Runner, such as demography or face feature extraction. It may also be used for functional filtering but Detection Filter below is recommended for that purpose, as it offers a configuration tool as well. The filtering criteria are

  • Detection confidence: OBJ_FILTER process node
  • Head size: OBJ_FILTER process node
  • Region of Interest (ROI): POLY_ROI_FILTER process node
  • Head Pose: HEAD_POSE_FILTER process node

For more information on the configuration of these nodes, see the MGR configuration format proto file.

Detection filtering by the Detection Filter microservice:

The purpose of this filtering is functional: it can restrict the detections according to the functional requirements of the application. The filtering criteria are

  • Detection confidence
  • Region of Interest (ROI)

A Detection Filter is made up of a single or multiple areas defined by polygons. Each polygon can be positive or negative, meaning detections in the polygon are included or excluded by the filter. This allows for precise and versatile filtering. If no positive areas are set, the whole screen is the searched area.

Polygons can be configured manually or drawn in the Stream Configuration UI for easier configuration.

For each Detection Filter a confidence level can be defined, providing another layer of filtering options.

Each Detection Filter runs in a separate microservice. If multiple filters are needed, additional microservice instances are required.

When Detection Filtering is enabled, only detections that fall into the specified area and reach the set confidence level, are recorded.

For more information, see Detection Filter.

← Skeleton DetectionTracking →