Study On Driving Interest Area Detection Method Based Orn Deep Learning

Nowadays people pay more attention to safety driving because of frequent traffic accidents which caused by the increasing number of vehicles and bad driving habits of some drivers in the process of driving.As a potential solution,the intelligent driving technology will become a research hotspot.At the same time,as a key technology of intelligent driving,driving interest area detection method has an important impact on the performance of intelligent driving or intelligent early warning system.Driving interest area refers to the area of the road ahead of the vehicle.It can provide road environment information for intelligent vehicle.At present,commonly used detection methods of the road can be roughly divided into two categories,one is to use features of specific road line for road detection,establish road model in advance,and conform the model constraints,then use marker characteristic to detect results and calculate model parameter.The other is to focus on the segmentation of the road area,and the classifier is trained by extracting the image color,texture,gradient and other features.However,due to the influence of many factors,such as the diversity of the lane,the illumination change and the shadow,most of the existing road detection methods are not robust enough.In view of the shortcomings of the existing methods,this paper proposes an effective method for driving interest area detection based on the deep learning.Firstly,we obtain the depth features from the SSD model training,and apply the extracted depth features to random forest training which can get the accurate horizon line estimation.Secondly,we obtain the camera internal and external parameters by using camera self-calibration method based on camera model.Thirdly,we use the Canny edge detection and K-means clustering to realize the vanishing point estimation.Finally,we establish the road detection model which based on the obtained estimates,and further realize the segmentation of road surface.In order to verify the feasibility and effectiveness of this method,we conducted the horizon detection experiments on the dataset KITTI and Guangzhou_ADAS.Detection results show that the proposed method not only can improve speed and accuracy in calculation,but also can provide stronger robust feature extraction method compared with the existing methods.At the same time,we mainly carry on the road detection on the KITTI data set.The experimental results show that compared with the existing methods,the proposed method not only has better road detection effect,but also can detect the road area of the shaded part more accurately.