Research And Implementation Of Fatigue Driving Detection Algorithm Based On Facial Multi-feature Fusion

Fatigue driving is the main factor affecting traffic safety,and it has great damage to life and property.In order to ensure the safety of drivers and pedestrians and reduce traffic accidents,it is of great application value to research and implement a vehicle-mounted fatigue driving detection system.The determination method based on the driver’s facial visual information has the characteristics of easy collection,accuracy,and no contact,and has great development potential and application prospects in various types of fatigue driving detection.Affected by factors such as head posture changes,partial facial occlusion,randomness of lighting conditions,and diversity of fatigue performance during actual driving,there are still many technical bottlenecks in all-weather and highly robust fatigue driving detection.This thesis focuses on the driver’s eyes and mouth features to identify the fatigue state,develop a fatigue driving detection algorithm,and design and implement software and hardware.In order to improve the accuracy of facial feature point location in large pose and partial occlusion when locating eyes and mouth regions,this thesis proposes a video facial feature point location and tracking algorithm based on multi-view constrained cascade regression.The method estimates the initial shape using the transformation relationship established by 3D and 2D sparse point sets,and uses affine transformation to perform a pose correction on the face image.When constructing the shape regression model,the left face,front face and right face are respectively used to construct the cascade regression model.Finally,a re-initialization mechanism is adopted to establish the shape relationship between consecutive frames using normalized cross-correlation template matching tracking when the feature points are located correctly.Based on the feature point localization results of eyes and mouth,this thesis combines the feature parameters of eyes closed and yawning,and proposes the use of weighted sum to achieve the classification of fatigue state levels.For the determination of eye state,this thesis proposes a method for human eye state discrimination based on the improved Mobile Net-SSD network.At the same time,in order to improve the accuracy of eye state determination when the pupil is partially occluded,an auxiliary determination of eye opening angle is introduced.For the judgment of the mouth state,the commonly used mouth aspect ratio is used to measure the opening degree of the mouth,and the continuous frame number of the mouth opening to a greater extent is used as the judgment feature of the yawning action.Aiming at the specific realization of the fatigue driving detection algorithm,this thesis designs a fatigue driving detection system combining software and hardware.In order to detect the driver’s fatigue under nighttime conditions,the system uses a near-infrared camera to obtain the driver’s face image.In order to improve the real-time performance of the system,the system adopts the NT96580 chip with deep learning acceleration function as the core processor,and designs the software system on this basis.Finally,the test experiments show that the system has high accuracy and real-time performance during the day and night,and the accuracy of the classification of fatigue status is above 90%,which has a certain use value.