Research On The Detection Technology Of Driving Behavior Hidden Danger Based On Multi Information Fusion In Complex Background

In recent years,traffic accidents caused by illegal driving behavior occur frequently,which always threaten people’s life and property safety.Using computer vision technology to detect the hidden danger of driving behavior has the advantages of good real-time,non-contact and so on.It has become an important method in the field of driving behavior safety detection.It has important academic significance and application value to carry out in-depth research on related technologies and engineering applications.At present,the detection of driving behavior safety hazards has the disadvantages of low real-time and small detection range.Considering the accuracy of detection and the anti-interference ability of detection process,this paper focuses on the low illuminance image enhancement algorithm,target detection and tracking algorithm and multi information fusion algorithm for the detection of driving behavior safety hazards from the perspective of improving the recognition rate and speed of driver’s smoking behavior and safety belt use behavior.The main research contents are as follows:(1)In order to solve the problem that it is difficult to process the video image when the light in the cockpit is insufficient,this paper studies and improves the low illumination image enhancement algorithm based on Retinex.The color space of the image is converted to HSV space,and the illumination component of V channel is estimated by the improved Retinex image enhancement algorithm,and the result is replaced by V channel;finally,the HSV space is converted to the original color space.Through the enhancement of low illumination image,the brightness,details and other information of the image can be improved.(2)Aiming at the problem that the smoke emitted by the driver is not easy to capture and the smoke feature is difficult to obtain,an improved vibe algorithm is proposed based on the research of vibe background modeling algorithm.The matching radius of the model is adjusted dynamically,and multi gradient threshold is used to update the neighborhood.The optimized algorithm can extract smoke accurately.Aiming at the problem of inaccurate facial information acquisition caused by the driver’s head movement,the CAMSHIFT moving target tracking algorithm is improved.In order to avoid selecting the initial frame of target tracking manually,a method of quickly locating the driver’s head position by facial contour and color is proposed.We use the vibe algorithm to find the suspected facial contour information,analyze the detected area,and determine the exact position of the face through the color features.By fusing Kalman filtering algorithm to improve CAMSHIFT target tracking algorithm,the improved algorithm can still successfully track the target when there is error in the predicted position of the target.(3)Based on the analysis of the existing target recognition and detection algorithms,a multi information fusion algorithm is proposed,which is composed of the driver’s mouth,the smoke in the cockpit,the driver’s seat belt and other video images.By extracting the behavior characteristics of drivers when driving a car,and building a classification model of driving behavior safety hazards,the driving behavior of drivers can be directly detected.Finally,this paper integrates the modules and evaluates the performance of the system.The results show that the algorithm is accurate and real-time.When the driver has dangerous driving behavior,it can directly prompt the driver and keep the current driving behavior information.