Human Detection and Target Location Estimation
In this post, we talk about Human Detection and Target Location Estimation.
The occupancy information is a crucial element in the Smart Localized Heating Control (SLHC) system. Thus, the image’s quality and type are also important. To attain image data in space, many different methods and advanced equipment can be used. In particular, the depth detection camera such as the time of flight (ToF) or the RGB-D camera is commonly used in the human recognition and movement detection projects. However, they are expensive and relatively high power consumption devices. The Microsoft Kinect (RGB-D) is popularly being used in the similar image processing projects because it is cheaper than the other depth detection camera. However, it is a still expensive (~$200 USD) for our heating system. Therefore, we decide to use a webcam that is a cost effective (~$25 USD) and has the capability to stream the real-time video images despite the image captured is only 2D images.
Due to the limited information of the webcam’s image, we use the integrating and utilizing method of the existing image processing techniques to obtain the useful level of occupancy information for the target’s coordinate in 3D space. use five distinctive methods listed below to get the result of human movement detection and target location estimation.
- Movement Detection (Human Movement Detection)
Detect the intensity change and segment the region
-
Human Verification (Human Movement Detection)
Detect the human features, face and shape
-
Human Height Determination (Target Location Estimation)
Capture the pixel values around the human image in 2D space and estimate the height to find the initial target point of the region
-
Distance Computation (Target Location Estimation)
Capture the lowest extrema point (foot location) and compute the distance between the webcam and the person
-
Target Point Determination (Target Location Estimation)
Determine the target zone based on the target coordinate in 3D (x, y, z) when z represents distance (d)
In order to accomplish the 3D target point detecting function of the system, we use the state-of-the-art image processing techniques such as human feature detections (face detection, histogram of gradient (HOG)) and the basic image processing methods (intensity segmentation, contouring, mass center detection, and background subtraction using the Mixture of Gaussian (MOG)). Also, develop the TLE algorithm to compute 3D coordinate value from 2D source image.