The Kinect can get both depth information and color information of the scene. For object detection, I integrated the color and depth information to a new map, and extracted HOG feature on this map for SVM training. On another application, I took the depth information from a vertical Kinect sensor as only resource, and proposed a novel unsupervised method which simulated the process of water flow to detect people’s head.
For the detection on public dataset, I have participated in several projects about face detection and pedestrian detection. We introduced hierarchical part based structural model for face detection, which contained part subtype option and part deformation. Then we presented an effective deformable part model for face detection in the wild, with faster HOG extraction, sparse constraint and cascade. On another hand, we proposed a Multi‐Task model for multi‐solutions pedestrian detection which jointly considered their commonness and differences.
Besides, I work as image algorithm engineer for some commercial projects provided to Chinese security department