Summary

We efficiently detected the taillights of front vehicles in nighttime environments by converting the RGB color space to HSV and YCbCr. The detection of taillights was used to generate a candidate pool for vehicles. Haar-like features and the AdaBoost algorithm were then employed, validating the vehicles using 2000 positive and 5000 negative images. A Region of Interest (ROI) was specified to reduce processing speed in real-time video processing.

To calculate the distance between vehicles detected based on taillights, we estimated the vehicle width based on each taillight and computed the inter-vehicle distance. To effectively communicate the risk of a collision to the driver, a Head-up Display (HUD) was created, indicating danger levels using three levels of sound (Forward Collision Warning - FCW).

To further reduce the error rate in vehicle detection, we added tracking using Mean-Shift Tracking or Kalman Filter to track future video frames of vehicles. Additionally, for improved performance in forward vehicle detection, we plan to explore the use of a more accurate classifier continuously. Our future research will focus on exploring effective methods to enhance performance continuously.