Research On ADAS Intelligent Image Recognition And Ranging Technology Based On Deep Learning

Nowadays,with the rapid economic development and the increasing number of vehicles,road traffic safety has gradually become a major problem.In order to reduce the traffic accidents,ADAS(Advanced Driving Assistance)system is being applied by more and more automobile manufacturers.The most important part of the collision avoidance system in ADAS is its recognition and ranging part.The complexity of road traffic environment and the diversity of pedestrians and vehicles bring great difficulties to the vision-based recognition and ranging system.Therefore,it is important to study the vision-based recognition and ranging system and improve its ranging accuracy.In view of the fact that the traditional algorithm works well in detecting the target,but the positioning accuracy is poor,which greatly increases the error of the ranging system,also greatly compromising the security of the entire ADAS system.In this paper,a new recognition and ranging system is proposed.The algorithm consists of two parts:detection and ranging.The detection part is based on convolution neural network,and the ranging part is based on vanishing point detection and monocular vision.The emphasis is to improve the overall recognition and ranging accuracy.In the part of ranging algorithm,several classical monocular ranging algorithms arc studied and analyzed in this paper.It is found that the classical ranging algorithms are not practical enough to be directly applied to the actual scene.Based on the particularity of the road driving environment,an algorithm based on the location the vanishing point which is the intersection point of the road edge extending in the image.In the part of detection algorithm.firstly,this paper analysed and introduced the traditional detection algorithm in dctail,and then introduced the detection algorithm based on deep learning.Also introduced the advantages and disadvantages of traditional recognition algorithm and detection algorithm based on deep learning are analyzed.Finally,a network which first extracts the detection frame from the rough image detection,and then precisely segmentes the image is proposed.This makes the network not only improve the detection accuracy,but also reduce the computational load of the network.Finally,the test is divided into two parts,in the detection part.This paper uses 100 test-set pictures in sunny day,rainy day and night to test detection proposed in this paper and other detection algorithms,and obtains the recognition accuracy,location accuracy and calculation time consumption,and analyses the advantages and disadvantages of each algorithm.For the ranging part,this paper uses a total of six fixed distance pictures than use the ranging algorithm proposed in this paper and other ranging algorithms for ranging,obtains ranging errors at different locations,and analyzes the advantages and disadvantages of different algorithms.This paper also tests the influence of location errors of different detection algorithms on the ranging accuracy of ranging algorithms.Compared with the traditional algorithm,the detection accuracy and ranging accuracy of the recognition and ranging algorithm in this paper are much higher,and the change of the environment and weather has little impact on the algorithm.At the same time,the algorithm in this paper can maintain an acceptable real-time with high accuracy.