| With the rapid development of economy,vehicles have become an indispensable part of people’s life.However,the resulting parking difficulties are becoming more and more prominent.As the traditional parking space guidance mode used in large urban parking lots is not highly accurate and the display of empty parking spaces is not clear.It is inconvenient for drivers to find parking spaces.They even cannot get the information of empty parking spaces accurately and quickly.In order to enable drivers to obtain parking space availability information in real time and accurately,and save the time consumed in finding parking spaces,thus solving the parking difficulty problem.In this thesis,we studies an improved vehicle detection method based on YOLO algorithm,research and design a parking space guidance system based on the improved vehicle detection algorithm.The specific work is as follows:(1)In order to solve the problems of high cost,difficult maintenance and low recognition rate of small target vehicles and obscured vehicles in existing parking lot vehicle detection methods,a vehicle detection algorithm based on improved YOLOv5 s is proposed.Firstly,we added a self-attentive mechanism module(Convolutional Block Attention Module,CBAM)to improve the model’s focus on salient features,which in turn improves the detection accuracy.Secondly,we added a dual-scale feature fusion target detection technique(Bidirectional Feature Pyramid Network(Bi FPN),which reduces the number of non-essential nodes in the network structure and improves the feature extraction capability of the model.These methods improved the detection accuracy and speed of the algorithm.(2)In order to make the vehicle detection network more flexible and better able to be applied to actual complex scenes,to meet the demand for speed and accuracy of vehicle detection of small and obscured targets in different scenes,the faster and more flexible YOLOv7-tiny algorithm is used.Based on this algorithm,the ACmix structure is added to improve the feature extraction ability of the network and enhance the capture of information features by the network,thus improving the recognition and detection ability of the network structure for small targets.At the same time,in order to adaptively learn the feature correlation between different levels,the Adaptively Spatial Feature Fusion(ASFF)network is introduced,which is an adaptive spatial feature fusion network.Fusion(ASFF)is introduced,which fuses the feature maps of different levels by learning the weight parameters,and then enhances the detection of small targets as well as obscured targets.(3)Based on the improved vehicle detection algorithm,this thesis studies and designs a real-time parking space guidance system,which realizes the parking space guidance function by dynamically displaying the number of empty parking spaces and their lanes.After the test results show that the system can reach 99.1% of the correct detection rate of the parking space.It can well complete the task of parking space guidance,meet the demand for detection accuracy and speed in the actual complex scene with good detection effect. |
