(English) People Detection SUP Plugin

{supProcessFilterPatchObjectDetection}

Interest: Our purpose is to be able to optimally detect people, faces and other body parts in video sequences.

Method: For feature selection we use a variant of LBP Features designed by (1). Adaboost is used for training the data and a cascade-model for detection.

Challenges: In most of the sequences we work with, generic S.o.t.A. detectors would fail due to the following variables:

– distortion due to the Lens and/or Point Of View of the camera

– low resolution, low FPS

– light changes

– occlusions

As an example, in some of the video sequences people appear tilted as they get further from the centre of the scene and are missed by the regular detector.

{supProcessFilterPatchObjectDetection}

{supProcessFilterPatchObjectDetection}

{supProcessFilterPatchObjectDetection}

Previous contributors: Étienne CORVÉE , Vasanth BATHRINARAYANAN

Current contributor: Silviu-Tudor SERBAN