Vehicle Tracking in Video Based on Pixel Level Motion Vector - Multimedia and Signal Processing - page 202

3 Pixel Level Motion Vector

To solve the problems of losing vehicle blocks in block level motion vector based
tracking, this paper provides an vehicle tracking algorithm based on pixel level mo-
tion vector to ensure that the vehicles can be tracked farther. In this algorithm, an
image of 640 × 480 pixels is divided into 633 × 473 blocks, where each block consists
of 8x8 pixels, namely, each pixel at the left and top region consisted by pixels
corresponds to a block. The flow chart is shown in Fig. 2.

Fig. 2. Flow Chart

3.1 Detecting New Vehicles

Set up a detecting area on the entrance of the scene as shown in Fig.3. It is assumed
that all vehicles will come through the detecting line. Therefore, we can assign new
vehicle IDs only to the foreground regions in the detecting area. The foreground re-
gions can be obtained by calculating the difference between the k th frame image and
the background image of the input video. A connective region where each pixel has a
larger difference than a threshold value is regarded as a foreground region.

Fig. 3. New blocks of a vehicle in the detecting area