Figure 12. FVT system architecture
According to the vehicle position identified by FVD, the segmented ROI will be presented in next frame. The size parameter of the segmented ROI is same with the outcome of FVD. After that, the binary image to be erosion and dilation and CCL, and computing vehicle position, then compare to current vehicle and vehicle position of previous frame. Due to prevent from vehicle tracking error, calculate the current vehicle of number of occurrences is more than five times. If the result is true, FVT running continuously, else redo the road scan line to detection.
IV. EXPERIMENTAL RESULTS
The proposed algorithm is applied to TI DM648 DSP platform and running on the actual road with various environments. The input image format and size is with PAL 720x576 pixels and frame rate is 30 per second.