ParKing Slot Detection Under The Car Surround System

Parking slot markings detection is one of the key steps of the automatic parking system.This paper proposes a vision-based visual parking space detection system,and identifies obstacles in the parking space,which can realize empty parking space detection more safely and accurately in complex scenes.The main research results of this paper are as follows:(1)In order to capture a panoramic image of the surroundings of the vehicle,in this paper,a low-cost bird’s-eye vision system is constructed by installing four fish eye cameras around the vehicle to provide an image covering the surroundings of the vehicle.In addition,we designed a special calibration cloth and proposed a secondary calibration algorithm to correct the distortion of the fisheye lens.Then the experiment proves that the secondary calibration result is more accurate than the first calibration result.(2)A method for automatically detecting parking spaces is studied.The method is mainly composed of separation line detection and parking entrance detection.In the separation line detection stage,we propose a new line clustering algorithm based on the LSD(Line Segment Detector)algorithm;In the parking entrance detection phase,we propose a learning-based approach that uses LBP(Local Binary Pattern)features,Haar(Haar-like)features and AdaBoost(adaptive boosting)classifiers to train T-or L-shaped marking points at the entrance to the parking space,and the convex defect algorithm is used to classify the marking points.Due to the current lack of parking space data sets,we propose a multi-visual fusion method to increase the number of training sets.Finally,the parking space occupancy is determined by calculating gray histogram.Experiments show that the accuracy and recall rate of the algorithm are over 97%.(3)In order to detect the parking space directly in the distorted fisheye image,this paper proposes a real-time parking space location algorithm based on semantic segmentation.The method is a fully automatic method,and can detect various shapes of parking lines under different illumination conditions in different application scenarios.Firstly,we use deeplabv3+network to segment the parking line and get the segmentation result.Next,we use the improved YOLOv3 algorithm to detect the obstacles of the pillars and vehicles in the parking lot.Then,the results of semantic segmentation and obstacle detection are combined to obtain effective parking spaces.Finally,we evaluate the proposed algorithm by plotting the Loss curve and the IOU curve and the two evaluation indicators Recall and Precision,and verify the effectiveness of the algorithm.