Research On Campus Parking Space Occupancy Detection Algorithm Based On Vision

With the growth of China’s car ownership,the pressure on urban traffic is heavy,and the intelligent development of urban traffic is promoted.Accurate and efficient access to available parking spaces is critical to solve the increasingly difficult parking problem.For the detection of parking spaces occupancy,traditional parking lots usually have geomagnetic coil sensors buried in the ground.When the vehicle comes in or leaves,the parking space occupancy is detected according to the signal of the geomagnetic coil sensor.Although this method is feasible,geomagnetic coil sensors may be affected by factors such as complex physical environments.Such as on-street parking spaces,it is difficult to bury geomagnetic coils on the road surface and the maintenance cost is high.With the development of computer vision,a variety of vision algorithms have matured.Capture image frames through cameras and use computer vision algorithms to analyze parking space occupation,which is not only efficient,but also saves costs.At the same time,the physical environment in the parking lot can be monitored by cameras.Therefore,this paper proposes two vision-based parking space occupancy detection algorithms,as follows:(1)This paper proposes an integrated classification parking space detection algorithm that fuses image texture and shape features.Parking space occupancy detection can be understood as the classification of images,that is,the classification of vacancy status images and images occupied by vehicles.For image classification,traditional methods extract single features of the image,such as color features or gradient features,and then send the features to the classifier for training to produce detection results.This paper considers the diversity of features and fuses the texture and shape features of the image.Extract LBP features in blocks,extract Sobel features,cascade the two feature vectors,and apply principal component analysis to remove redundant features;Then,the Adaboost integrated classifier is used to classify the parking space image.The experimental results show that the evaluation indexes of the algorithm is higher than that of the single feature algorithm.(2)A convolutional neural network based on non-local operations for parking space occupancy detection algorithm is also proposed.Aiming at the disadvantages of poor generalization performance of the traditional artificially designed algorithm,this algorithm is based on a convolutional neural network.However,the convolution operation is not flexible enough for the transmission of long-distance information of the image.Therefore,we introduce a non-local operation to directly obtain the high-frequency features of the edge by measuring the similarity between long-distance pixels,at the same time,a small convolution kernel is used to obtain local detailed features,and the network is trained in an end-to-end manner.Set different convolution kernel sizes and the number of non-local module layers to optimize the network structure.The experimental results show that the proposed algorithm has significant advantages over traditional parking space occupancy detection algorithms based on artificial design features,both in accuracy and generalization performance of the model.Compared with the widely used convolutional neural network,this algorithm has greater advantages.In the real scene,based on the internet protocol camera,we obtain the parking space image through the Open CV software platform.And we use Docker and Tensor Flow Serving software platforms to deploy the algorithm model.In the tests of the two parking lots,the model achieves a high accuracy.This paper aims at the problem of parking space occupancy detection,and summarizes the progress and shortcomings of existing research.From the perspective of image recognition,this paper proposes two visual parking space occupancy detection algorithms,one is a fusion of texture features and shape features,and the other is based on a non-local operation convolutional neural network.Good results have been achieved in practice.