Research On Real-time Vehicle Type Recognition For Intelligent Vehicle

Intelligent vehicles have two meanings: Intelligence and ability.Intelligence is refer to the capability of a car to perceive,judge,reason,decide,and remember as intelligently as human beings.Ability is the capability of intelligent vehicles to ensure the effective implementation of intelligence.Autonomous driving is the organic combination of Intelligence and ability,which is complementary and indispensable.Autonomous driving technology generally includes environmental perception,decision planning and vehicle control.As the first step of autonomous driving,environmental perception is in the key position of intelligent vehicle and interaction with external environment information.With the development of computer vision technology,more and more scholars have studied vehicle identification based on video image.Given that intelligent vehicles need real-time perception of surrounding environment information during driving,different control strategies are adopted according to the various of vehicles detected.Accurate and real-time detection of various obstacles around the vehicle is an important indicator for autonomous driving technology,which is crucial for autonomous vehicles.This paper is positioned in the real-time vehicle identification of environmental perception and about optimizing and improving the algorithm for the low detection rate and poor real-time performance of vehicle targets in environmental perception.The main work of this paper are as follows:First of all,this paper introduced the current research status of intelligent vehicles and the current research status and difficulties of vehicle identification system.Based on the comparison between traditional visual recognition algorithm and deep learning recognition algorithm,this paper analyzed the characteristics of traditional machine learning algorithm used by vehicle identification and the process of vehicle recognition algorithm based on deep learning algorithm.Secondly,based on the convolutional neural network,an end-to-end model identification algorithm for intelligent vehicles is designed and the model database used in this paper is constructed.Based on this database,vehicle detection algorithm between traditional machine learning and deep convolution neural network are compared and analyzed.The average accuracy of vehicle detection algorithm designed in this paper is ten percentage points higher than that of traditional machine learning algorithm.Then,this paper introduced the algorithms for the compression and acceleration of the convolution neural network model,which is optimized by using the model compression and acceleration technology for the slow problem of the feature extraction of the convolution neural network.The lightweight network architecture is introduced into the design of the convolutional neural network,which reduces the size of the network model and effectively speeds up the reasoning speed of the model.Finally,based on the Drive PX2 computing platform released by NVIDIA,this paper verifies the algorithm on the vehicle type database.Compared with the traditional visual recognition algorithm,the model recognition method designed in this paper achieves the desired effect.In this paper,the design of vehicle detection algorithm based on convolution neural network fully considered the characteristics of different levels in the network information.Under the premise of ensuring the real time,the missing rate of vehicle target detection is reduced effectively,which laid a foundation for the application in the field of automatic driving.