Key Technologies Research Of Advanced Driver Assistance System

With the rapid development of the human society and the automotive industry,the automobile became more and more smart.Traffic safety is a problem that every driver must pay attention,and it makes how to improve the automobile active safety to become a research hotspot.Automobile safety driving assistant technique been an effective step to reduce traffic accidents and accident losses,and it is the developing trend of automobile in the future.Advanced Driver Assistant System(ADAS)allows the driver to rapidly detect the possible danger.This system uses various sensors which installed on the automobile to collect environmental data inside and outside,and it can detect and track the static or dynamic objects.The ADAS for a certain electric automobile is researched in this paper,which used the digital camera as the sensor to collect the image around the automobile.The system is set up with the core processor-iMX6.The key technologies are studied,such as the lane departure warning system and the pedestrian recognition system.The main research contents consist of the following three parts:(1)In the image denoising preprocess,a new denoising model based on new diffusion coefficient and setting threshold parameter K automatically is proposed in this paper.In this model,the diffusion coefficient is determined by threshold K.If the coefficient is less than K,it has the ability of diffusion and is expressed by a function accordingly.And if it is greater than K,it is 0,thus limiting the influence of the isolated points caused by noise on the image quality.The threshold parameter K is determined by the gradient of each iteration.The value of the gradient is determined by the 8 directional values of adjacent pixels in the pixel position.Finally,the improved diffusion coefficient is plugged into the PM model to realize the denoising.The experiments show that this model reduces the noise while preserving the edge features of the image.Therefore,the new model is better than the existing PM model and NLM model in terms of signal to noise ratio and structural similarity.(2)In the lane departure warning system,an improved method aiming at the structured road and clear lane line without vehicle occlusion is discussed in this paper.This method is based on the existing linear Hough transform algorithm,and it combines the thoughts of LMedsquare and least square method to perform lane fitting.This method improved the recognition rate of lane,meanwhile it adopts difference methods to track the lane.According to the position relationship between the automobile and the lane image,the distance between them is calculated,and an early warning decision model is established to judge whether the warning occurs.The experiments show that this method can improve the accuracy of lane recognition and meet the real-time system's requirements.(3)Because the existing pedestrian detection methods don't have nice performance in the real-time system,a feature extraction method based on Leg-Histogram of Oriented Gradient(L-HOG)is proposed.The real time problem is solved by selecting the appropriate feature region to reduce the feature dimension.Then the SVM classifier can be used to subdivide the pedestrian sample into the front and rear view one or the left and right view one.It can be used to judge the pedestrian's movement state through the leg posture.The experiments show that the method's calculating speed is increased while the recognition rate of pedestrians can be guaranteed.