Specifications

possible to 10 feet. Only a single video frame is needed with the un-occluded markers for the

calibration. The markers need not stay in place for the entire duration of the video. The

calibration process can be completed at any time during the video recording as long as the

camera remains fixed throughout.

Once the calibration process is completed, a conversion factor between distance in pixels in the

camera image and distance along the calibration plane in the scene can be determined. However,

the factor resulting from this calibration is only an approximation. Deviations from this

conversion factor are expected when the tracked objects do not lie on the calibration plane (i.e.,

the objects are closer or farther), or when the tracked objects are on the calibration plane, but

distant from the location of the markers (i.e., just entering or leaving the image frame). For the

purposes of vehicle tracking in the lane of interest, these deviations are minimal. A calibration

process involving more markers placed at a range of known camera depths could be used to

minimize this error if needed for a particular application. This approach was avoided in order to

make usage of this tool as simple as possible without introducing an unreasonable amount of

error.

4. DIGITAL VIDEO ANALYSIS

The digital video analysis begins with a frame by frame extraction of objects in the scene. These

objects are extracted by comparison of each frame with an image of the scene that does not

contain any objects of interest (i.e., vehicles). Then, a position-based criterion is used to link

objects across frames in order to create object tracks. The object tracks are analyzed to determine

object identity (i.e., vehicle, pedestrian). A dynamic occlusion detection and correction analysis

is conducted in order to uniquely identify each object in the scene and correct the estimated

object center of mass. The output of the analysis is a dataset containing the position, velocity,

size, and identity of each detected object in the scene. The following sections describe the steps

of this process in more detail.

4.1 Object Detection

The output of the object detection stage is an object detection list. This is a list of all detected

objects in each video frame that specifies the position and size of each object. Because a single-

camera approach is taken in this research, it must be assumed that the detected objects are on or

near the calibration plane. Thus, the positional information is limited to horizontal and vertical

position estimates, and the size estimates of the vehicle refer only to the side profile of the

object.

Each frame of the recorded digital video is extracted to a separate image. Each image is then

point-wise compared to an image of the scene that does not contain any objects of interest. This

image is referred to as the background image. Prior to proceeding with object detection, the user

should select a single frame from the video that does not contain any objects of interest or any

other moving objects. While an automatic background selection algorithm is available to the

user, the algorithm is relatively slow and will not guarantee that the image does not contain any