动态光照下驾驶人面部特征识别算法与试验研究

Algorithm and Experimental Research of Driver's Face Feature Recognition under Dynamic Illumination

基于机器视觉的驾驶人面部特征识别受光照的影响很大。为克服由于动态光照引起的背景干扰，面部特征弱化的问题，采用一种基于KalmanFiltering的光照自适应AKF算法，通过高斯概率密度函数建立Gi（i，j）算子，实现驾驶室背景的分割；在HSI色彩空间中通过闽值分割算法提取面部肤色区域，最终建立了眼鼻坐标搜索模型；进行了不同的照度与头部姿态下的AKF—HSI算法试验，测试统计前景分割率Kfrontground、肤色分割率kski．与眼鼻识别率δ，在2×10～10×10^4lx的照度下，眼鼻的平均识别率艿达到82％-92％。结果表明AKF—HSI融合算法对动态光照下眼鼻识别具有较好的鲁棒性，照度E、头部姿态与硬件设备AGC是眼鼻识别的最主要影响因素。

The recognition of driver＇s facial features based on machine vision is influenced greatly by illumination. In order to overcome dynamic illumination induced background interference and weakening of facial feature, a self-adaptive AKF algorithm based on Kalman filtering is adopted. Real-time segmentation of driving cab background is realized by Gi （i, j） operator established by Gaussian probability density function. By clustering performance of skin color and facial feature in HSI color space, skin region and facial feature are recognized and located by the threshold segmentation algorithm. Finally, an eye-nostril search model is set up by their coordinate value sequence. The AKF-HSI algorithm experiment is then carried out under different illuminancies and head poses. Frontground segmentation rate kfrontground, skin segmentation rate kskin and eye-nostril recognition rate 8 of different persons are obtained, which shows that the average recognition rate 8 reaches up to 82% - 92% with the outdoor illumination of 2 × 104 - 10 × l04 lx. The result shows that the anti-interference performance of AKF-HSI fusion algorithm is robust to eye-nostril recognition under dynamic illumination. Illuminancy E, head pose and AGC hardware equipment are the most important influence factors of eye-nostril recognition.