|
Introduction: |
|
The alarm system
algorithm is implemented as an automaton with three basic states (or
“stages”). A short presentation of the main implementation issues for each
of these stages is given in the following paragraphs. |
|
Stage 1: |
|
Recognition of two objects of different sizes, which are placed in the field of view of the system.
- A set of special image filters are applied to eliminate the noise introduced by the video camera:
median filter, average filter, dilation and erosion.
- Motion detection is based on the frame differentiation technique, using a special approach: a circular buffer
of 20 image frames is used to store successive frames. When motion is detected, the current frame is also
compared against the last frame in the buffer (which is considered the “original frame” at the time).
- Differentiation between the two objects is based on the calculation of the area of the movement detected by
the algorithm.
- When the system detects enough movement to identify the largest object, the Level 2 alarm is issued automatically.
|
|
Stage 3: |
|
Detecting person motion in the protected area and the attempt to remove the protected object
- First, the object to be protected is detected and identified by the system, once it is placed on the table and the system is enabled. The main image processing techniques used are image segmentation (Reddi algorithm with a variable number of thresholds) and edge detection routines (Sobel algorithm).
- The system accepts and detects any number of objects that are placed on the table, as long as they can be distinguished with this particular video camera (they feature good contrast with the table).
- Person movement is detected using a similar frame difference approach as presented at Stage 1, with a frame buffer of 10 frames.
|
|
Stage 4: |
|
Protect an area and a particular object inside the area, with motion on the background
- This stage is similar to the previous one, with the main difference that the image area corresponding to the background movement is removed and ignored during processing.
|
|
System Calibration: |
|
Special system calibration procedures were developed to compensate the lighting conditions that differ significantly in the laboratory (the environment in which the application was developed) as compared to the contest area.
Further on, an automatic calibration procedure was developed, to compensate the noise induced by the video camera: 20 consecutive frames are sampled from a static scene and differentiated. The result indicates the maximum noise level induced by the camera on a particular environment. |