Research On Driver 's Face Detection And Tracking Based On Infrared Video

With the development of road traffic conditions, people’s life becomes increasingly convenient. However, the increasing number of cars leads to traffic accidents frequently. One of the important reasons of traffic accidents is the fatigue driving. Therefore, the research of fatigue detection system is of great significance. Among the various fatigue detection system in the present research, the system based on video analysis has been paid more attention because of its advantages of non-contact and practicability, in which face detection and tracking are basic and important in the process of fatigue detection.The face detection method based on the Haar feature and the AdaBoost algorithm is utilized to scan the source image with the adaptive moving step, therefore the number of non-face windows is reduced. As a result, the detection speed is improved under the assurance of the face detection accuracy. A fast compressive tracking algorithm combining feature selection with secondary localization (FSSL-CT) is proposed to track the detected face. Firstly, the compressive sensing theory is utilized to extract the compressive features in each sub-region partitioned from the global region. Then, based on the distribution difference of the compressive features between the positive and negative class, some compressive features with strong classification are selected to construct classifier. The variation degree of the features distribution between the two adjacent frames is regarded as the learning rate, and the update speed of the classifier model is adjusted based on the learning rate and the positive class threshold. Finally, the secondary localization strategy is utilized to track the target. In each localization procedure, candidate samples are collected in the corresponding search region first, and then particular numbers of compressive features are selected based on their classification ability, and the candidate samples are classified by the classifier at last. The target is tracked after the two localization procedures.The experimental results on the public test sequences and the self-made infrared test sequences show that FSSL-CT algorithm performs favorable abilities in terms of tracking success rate, precision and speed under the conditions of short-time occlusion, illumination changes and pose variety.