Driver Face Detection Using Multi-feature And Multi-clue

As the Trafic Safety issues attract more and more attention in our society. The deriver face detection technology becomes a very important module in the Intelligent Transportation System. For the special environment in the cab, normal face detection methods all get poor performance. To figure this problem, this article uses multi-features and multi-clues for driver face detection. We get perfect pervormance on real vehicle front view image data set, whitch is collect by surveillance on the road. The prediction rate and recall rate are both more than 90%.The driver face detection architecture we designed consists of three stages:In the first stage, we use vehicle location information which is got from vehicle detection, to extract detection region from the original image.Then combine the Aggregate Channel Feature(e.g. LBP, HOG etc.) and Boosting based detection method to detect the driver face from the detection region obtaining some candidate windows.Finally, applying the idea of part-based model, we regard the lience plate and driver’s face as a whole entirety, and verify the rationality of the parts by verify the rationality of the whole entirety.Besides, in the second stage, we try to improve the recall rate, in other word, reducing the missing rate regardless of the prediction rate. In the next stage, we apply the part-based model to reject the unsatisfied driver face candidate windows. In this way, we improved the prediction rate to be ideal, and has very little effect on the recall rate.We have collected 1402 vehicle front view images, which contains two parts of day and night. We verified the effectiveness of our method, got prediction rate of 99.33% and recall rate of 93.87% in night images, and prediction rate of 96.28% and recall rate of 91.46% in day images.