Research Of The Train Driver Fatigue Detection And Recognition System Based On Facial Feature

With the great development of the railway system and the continuous input of new technology equipments to the railway transportation safety, the workloads of the train drivers constantly strengthen. Besides, the unreasonable timetable and work environment of the train drivers will easily cause them out of order, doze off, and fatigue driving. Because of all these factors, fatigue driving is becoming a great safety problem in the railway transportation. Therefore, the guarantee and supervising system of the train drivers'working state have be paid wider concern and attention by the public.In this thesis, the author intends to develop a kind of the train drivers'fatigue detection and recognition system under the fundamental requirement:on-board, real-time, and non-contact. The purpose of this paper is to analyze the facial features of the fatigue driving train drivers in using the pattern recognition and the image processing knowledge, combining with the driving features and rules, study the appropriate fatigue detection and recognition algorithm of the train driver under the complicated illumination and the high frequency but low amplitude condition, and to embed the fatigue detection and recognition algorithm to the hardware platform which centered by the DSP-digital signal processing chip. The development of the fatigue driving detection and recognition system can be divided into three steps:Firstly, face detection algorithm:studies the Harr-like weak feature and Adaboost algorithm, finishes the training of the face machine classification which is based on the Adaboost algorithm, and puts forward the thoughts of adding Canny edge detection and partial face hit-percentage statistics to the face detection, which has achieved good practical results.Secondly, the development of the eye detection and state analysis algorithm:first of all, this algorithm is to train the eye detection machine classification which is also based on the Adaboost algorithm. Besides, a new kind of the eye state detection algorithm in using the regional feature comparison method is proposed, which is based on the analyzing the advantages and disadvantages of the Hough transform methods for circle finding and the gray projection method. With great robustness, this algorithm does not need precise geometric model, and uses the average gray value which based on the regional features. After repeated experiments and parameter adjustment, this algorithm has achieved high analyzing rate of the eye state.Thirdly, the DSP embedding of the fatigue driving detection and recognition algorithm: uses the DSP/BIOS real-time operating system to embedding the fatigue driving detection and recognition algorithm into the DSP. In the embedding process, for the using of many optimization methods, the fatigue driving detection and recognition algorithm finally has reached the detection speed of 18 frames per second, which satisfies the performance needs of the real-time detection.After the accomplishment of the development of the fatigue driving detection and recognition algorithm, two kinds of fatigue driving detection and recognition system are developed, which are based on PC and DSP. Then we analyze the accuracy, real-time ability, and the availability of algorithm. The result shows that the algorithm basically reaches the expected goal, however, there will be more works to do before the real application stage.