Research On Vision-based Vacant Parking Slot Detection Algorithm

With the continuous growth of the number of vehicles in China,the shortage of urban traffic resources is becoming more and more serious,and the problem of parking difficulties is plagued by more and more drivers.As the last link of intelligent driving technology,automatic parking systems are not only for passengers’ experience it has an important impact,and it is of great significance to whether the intelligent driving technology can finally be put into application.In the automatic parking system,the detection of parking slot is a very critical part,and the accuracy of parking slot detection directly determines the final effect of automatic parking.In this paper,based on the vision technology route,the detection algorithm of vacant parking slot is studied.The detection of vacant parking slot is divided into two parts: the location of parking slot and the recognition of free parking slot.Combined with the characteristics of parking image,the location part is divided into two parts: the location of head corner and the reasoning of tail vertex coordinates.The main contents of this paper are as follows:(1)Aiming at the location of corner at the head of parking slot,this paper proposes a classification rule for corner at the head of parking slot,which can extract parking slot information comprehensively,and uses the rule to train yolov3 to complete the detection of corner at the head of parking slot.By analyzing the characteristics of different types of parking slot and the information needed in the reasoning process of the vertex coordinates at the end of parking slot,a classification rule for corner points of parking slot is determined.Using this rule to detect the corner of parking slot,the corner coordinates of the head can be determined,and the parking slot information that can be used for the reasoning of the tail vertex coordinates can be obtained at the same time.The corner classification rule is applied to label and classify the data set of look around parking image,and the YOLOv3 network is built.The labeled data set is used to train the network model,and the corner detection model of parking slot is obtained.(2)Aiming at the location of the rear vertex of parking slot,the reasoning location of the rear vertex of parking lot is completed based on the parking slot information obtained by corner detection.Based on the parking information obtained by corner detection,the reasoning rules of the rear vertex of the parking slot are given.The coordinates of the rear vertex of the parking slot are determined by reasoning,and the parking slot in the look around parking image is located.The experimental results show that the parking location algorithm proposed in this paper has good positioning effect in a variety of road environment.(3)For the recognition of vacant parking slot,convolution neural network is built and data set is made to train the network,and the recognition of vacant parking slot is completed.The convolution neural network is built and the image data set including the head area of parking slot is made as the training data.The network is trained to get the parking slot state classifier model.Experimental results show that the proposed method can complete the task of vacant parking slot recognition with high accuracy.Finally,a complete vacant parking slot detection experiment is carried out on the look around parking image.The experimental results show that the algorithm can complete the detection of vacant parking slot with high detection accuracy.