Research On Driving Assistance Technology Based On Multi-sensors Perception System

Drivers of large-scale vehicles and special vehicles face many problems such as large blind spots in the vehicle vision,complex driving difficulty and intensive manipulation,poor mobility of vehicles in unfamiliar environments or narrow lanes,etc.These problems would cause driving accidents easily.Driver Assistance Technology based on a multi-sensors perception system can observe the blind spots of a vehicle’s body in multi-directions and multi-angles,assist driving,reduce the driving complexity,improve the safety and mobility on particular traffic roads.It uses multi-sensors perception and calculation,panoramic image stitching and evaluation methods of vehicles’ mobility.According to the functional requirements of the multi-sensors perception system of the large-scale vehicle,a multi-sensors perception system based on visual sensing,image stitching and information processing platform with FPGA and HI3559 is designed.The main tasks are to design and complete each sub-system,establish the coordinates of the installed sensor,obtain the scene of the detection area of the vehicle’s body within a range of nearly 5 meters,and build the platform of the experimental vehicles.It can also calibrate the internal and external parameters by combining the target and Zhang Zhengyou’s algorithm,and the repetitive error of the calibration is concentrated within one pixel.Design the processing flow of the software algorithm of the multi-sensors perception system to stitch the panoramic and multi-channel videos and images.The region of interest can be extracted by dividing the regions into blocks and analyzing the features of the splicing regions.The algorithm of stitching panoramic videos and images is optimized by replacing the FAST algorithm with the original algorithm of extracting feature points,using the machine learning algorithm to filter and obtain a practical set with feature points and implementing the algorithm of matching feature points and RANSC algorithm.The correct rate of matching images of this improved algorithm is 87.6%.A complete and clear panoramic bird’s-eye view is obtained;the output frame rate of the system is greater than 16 frames/s,which meets the requirement of observing panoramic view and real-time display of the large-scale vehicle.To improve the assistance with observing the information surrounding the vehicle for drivers effectively,an algorithm of flexibly switching the viewing angles is proposed.The observation of human’s eyes is simulated by setting the coordinates of a virtual camera.The model of the algorithm of flexibly switching the viewing angles is established by the transformation of the inverse projection in the bowl-shaped projection model.Based on the reading data of deflection angle of the driver’s sight or obstacles detected by millimetre-wave radar,the flexible switching and display of the viewing angles of the panoramic bird’s-eye view can be achieved by using OpenGL.The experiments for detecting the target in the close-range environment of the vehicle verify that it assists the driver to observe the blind spots of the large-scale vehicle.The road condition,lane boundary and obstacle can be detected based on the panoramic image,the image with switching the viewing angles flexibly,referring to the standard of the vehicle braking distance and marking the area where the large vehicles are allowed to drive with the identification lines.With the panoramic image and the parameters of vehicles,the parameterized model of the driving road can be constructed to obtain the information,e.g.,the interval of the steering angle of the vehicle.The identification line of allowing passing assists the drivers to detect the driving road and schedule the driving routes.The results of the large-scale vehicle driving on the narrow straight and curved roads indicate that increase the driving assistance system,the efficiency of mobility is improved by 27.8%and 52.5%respectively,while driving on the narrow straight and curved roads.It concludes that this method can improve the mobility of large-scale vehicles in particular condition of the roads.