| The era of intelligent driving has arrived.Visual perception with cameras as a perception tool is a key part of intelligent driving,which accurately perceives the environmental information around the vehicle and plays an important role in assisted driving tasks such as traffic sign recognition and lane departure warning.However,in low-light environments,vision sensors do not have enough access to road traffic information,resulting in degraded performance of intelligent driving systems.Therefore,vision-based road information enhancement and detection technology under low illumination conditions has become a research hotspot.This topic is a sub-topic of the major project of Guizhou Provincial Science and Technology Department(Project No.: ZNWLQC [ 2019 ] 3012),which mainly focuses on the characteristics and detection technology of road traffic information in low-light mountainous cities.The details are as follows:1)Taking urban roads in mountainous areas as the research object,the characteristics of urban road images in mountainous areas under low illumination conditions are analyzed in depth.Using the existing experimental conditions,low illumination road images in real scenes are collected and sorted out,and a special data set is established to provide data support for subsequent algorithm verification.2)Aiming at the problems of insufficient illumination and contrast,color distortion and noise increase in urban road images in low illumination mountainous areas,two schemes are proposed to enhance the quality of low illumination road images.Scheme 1 proposes an enhancement method based on threshold partition weighted brightness component,which processes the image in the spatial domain.The maximum entropy method is used to design a threshold to partition the brightness component,and the improved hyperbolic mapping function is used to enhance the image pixels and improve image clarity and contrast.The experimental results show that this algorithm can quickly improve the clarity of road images by 1-2 times within 180ms/fps.Scheme 2 combines the characteristics of spatial and frequency domains and proposes an enhancement scheme based on hyperbolic tangent and homomorphic filtering.Firstly,the hyperbolic tangent function is used to compress the dynamic range of image pixels,balancing the overall brightness of the image.Then convert to the frequency domain and further enhance the image traffic information using homomorphic filtering using a 4th order Butterworth high pass filter.Experimental results have shown that the brightness and contrast of the road image processed by this algorithm have increased by 1-3 times,making the details in the image more abundant,and the processing speed can reach within 400ms/fps.3)Lane detection is crucial for the safety of autonomous driving and plays an important role in improving driving comfort and achieving adaptive driving trajectory planning.Therefore,a deep learning based lane detection algorithm is proposed to address the issues of low accuracy and poor robustness of existing lane detection algorithms in low illumination environments.By introducing residual networks and attention mechanisms,self built datasets are used for training and detection.The experimental results show that this algorithm effectively improves the accuracy of lane line detection and reduces the missed detection rate and false detection rate.Meanwhile,the average processing speed of this algorithm is154 fps/s,with a maximum speed of 300+fps/s,which can meet the real-time requirements under actual road conditions.By establishing a dataset of urban road images in mountainous areas under low illumination environments,we aim to enrich the low illumination scene image data in the field of traffic information detection.Design low illumination road image information enhancement algorithm and lane line detection algorithm,providing theoretical support for solving the computer vision research work of auto drive system under low illumination conditions. |
PDF Full Text Request