Research And Design Of In-car Lane Departure Warning System In Rain Based On Android

As one of the core advanced driver assistance technologies,lane departure warning system can improve driving efficiency and ensure drivers’ safety.Currently,it is deployed as an option on some high-end models.Also,it requires to be installed more components with complicated and expensive installation process,which is not conducive to the application of the system on low-end and mid-range models.In addition,the real-time,accuracy and application of lane line detection directly determine the overall performance of the lane departure warning system.However,in rainy conditions,rain can cause problems such as blurred and obscured lane lines in a way that reduces image quality.As such,it will bring about a significant reduction in accuracy of lane line detection,resulting in failure of the warning function.Furthermore,existing lane departure warning models cannot meet both accuracy and real-time performance due to its high complexity and computational burden.In response to the above problems,this paper works to design an Android in-car lane departure warning system for rain in light of the fact that many cars now have Android on-board central control screens.Also,this paper makes a lightweight design for the lane departure warning algorithm in rain and applies it to the on-board central control screen in the form of software.The main research content of this paper is as follows.(1)This paper has delivered beneficial outcomes in designing a modified light weight de-rain algorithm based on GAN.This paper provided key auxiliary information to address the adaptation of the model by introducing two learnable parameters,including ? and ?.Based on the generative adversarial network idea,proactive moves were taken to design a lightweight double-branch learning derain model called as DB-GAN from the perspective of feature-wise disentanglement.At the same time,endeavors were made in designing the generator and discriminator with a deep separable convolution.In terms of datasets,the Raindrop and Rain Land2023 were used to train and test the model in this paper.The experimental results showed that the DB-GAN derain network embraced low model complexity and high real-time processing task capability.Thus,it could play an active part in restoring the lane line details in the rain map.(2)This paper has registered steady progress in designing a lightweight algorithm for lane line detection based on an attention mechanism.This paper has designed a lightweight Ghost Net-CBAM-UNet algorithm for lane line detection in response to limited memory and computational resources in the on-board central control screen.Ghost Net V2 boasted the advantages of small number of parameters and operations.In this paper,Ghost Net V2 was chosen as the backbone network for lane line detection to perform a feature extraction on the input image and make an improvement in the UNet semantic segmentation network.Meanwhile,an introduction of the CBAM attention mechanism enhanced the expression of the network.Finally,efforts were made to fit the lane lines with an improved RANSAC algorithm and track them with the Kalman filter.The ultimate goal was to obtain stable and accurate fitting parameters for lane line.The experimental results suggested that the Ghost Net-CBAM-UNet algorithm for lane line detection could perform better in terms of accuracy and operational efficiency.(3)This paper has made greater improvements in designing and implementing a lane departure warning system for rainy days.Solid steps were taken to analyze the advantages and disadvantages of four commonly used lane departure warning models,establish an early warning decision model based on departure angle and lane lateral offset distance,and design a rain lane departure warning algorithm by combining a rain removal algorithm and a lane line detection algorithm.Using the ncnn forward inference framework,the model was transformed and deployed to the Android in-car central control screen in ways that complete the construction of the lane departure warning system in rain.It can be concluded that the system had an average memory footprint of around 120 MB,an average CPU footprint of 10%,a frame rate of 30.3fps and an alert accuracy of 93.85%,an improvement of 3.62% compared to the CCP model.Meanwhile,with good overall performance,the system could accurately detect lane markings in rainy conditions and give real-time lane departure warnings.