Research And Implementation Of Facial Image Fatigue Detection System Based On Gabor

With the development of the automobile industry, the demand for transportation is growing, and the traffic accident problem whose main reason is fatigue driving can not be ignored. Therefore, it is significant and urgent to perform fatigue detection algorithms reliably, fast and efficiently. However, previous works that identify fatigue based on the features of eyes and mouth can’t meet the real-time requirements. Furthermore, the researches have shown that the Gabor-based image feature extraction can capture multi-scale fatigue feature efficiently, multi-scale1-NN (1-Nearest Neighbor) classification algorithm can mine multi-scale feature information and obtain a good recognition rate, and the OpenMP based multi-core parallel computing can accelerate the detection speed efficiently. Therefore, in this paper, we proposed a new fatigue identification algorithm based on the ideas of parallel Gabor and1-Nearest Neighbor (1-NN) classification. The major contributions are three-fold as follows:1) To alleviate the complex computation for Gabor feature extraction, we proposed a parallel Gabor feature extraction algorithm. The algorithm is based on the OpenMP parallel mode, and it forks and joins the process of image Gabor transformation. Firstly, it divides the process of image Gabor feature extraction into partial parallel and non-parallel part. Then, multi-core computing is used to accelerate the process of partial parallel part. Finally, we combine the results. Experimental results show that the algorithm can effectively reduce the computing time of Gabor feature extraction and it can obtain a good speedup under mutli-core environment.2) Based on the mutli-scale feature that is extracted by Gabor wavelet image feature extraction algorithm, we proposed a mutli-scale1-NN algorithm to classify fatigue image. This algorithm classifies the image at different scales and integrates the results on all the scales to get the final fatigue classification results, based on the property that fatigue at different scales has different performances. Experimental results show that mutli-scale1-NN algorithm performs better detection than existing approaches.3) Based on the algorithms of parallel Gabor feature extraction and Multi-scale1-NN fatigue recognition, we designed and implemented a fatigue detection system using OpenCV2.0on VS2008platform, and performed tests on it. Experimental results show that our method can detect fatigue accurately and the performance of real-time fatigue detection has been improved greatly comparative to the single CPU computing environments.