The Research On Parking Space Detection Based On Magnetic Sensors And Convolutional Neural Networks

The number of vehicles in China has been increasing year by year in recent time,which has caused a series of problems such as parking problems,traffic jam and air pollution.Intelligent parking system effectively alleviates the problem of parking problems.The existing parking detection algorithms employ multiple technologies and sensors to detect vehicles,such as cameras,inductors,infrared,radio frequency identification and magnetic sensors.Thanks to the advantages of Magnetic sensors in terms of cost,endurance,scale,and deployment difficulty,many parking spaces now use magnetic sensors for vehicle detection.However,the existing magnetic detection algorithms all have some defects,which cannot handle the problem of base value drift and cannot cope with complex vehicle behavior.In this paper,we introduce deep learning into magnetic vehicle detection problems.We will use magnetic sensors to detect intensity of magnetic field,and use 1 D convolutional neural networks to classify magnetic signals.In this paper,we first propose a signal detection,acquisition and compression method based on magnetic sensors.The main idea is to use a magnetic sensor to detect the magnetic field environment around the parking space.The collected magnetic field intensity signal is preprocessed by a filter,and then we infer the start and end of the wave dynamics by an adaptive threshold,then the signal waveform of an event is determined,and the threshold is periodically updated according to the ambient noise.For the signal waveform within the wave dynamics,the magnetic field strengths and time information of the key points of the waveform are extracted.Then,the key points are subjected to differential quantization,so that the entire geomagnetic terminal compress of the original geomagnetic signal into a 38-byte compressed signal.Finally,the compressed signal is uploaded to the server for further processing.All the steps are ensured by the continuous MATLAB simulation to ensure the correction of the proposed method.Then,we propose a method of magnetic signal classification based on 1D Convolutional Neural Networks.The main idea is to use magnetic sensors to collect data and use the camera to monitor the vehicle behavior corresponding to each upload of data.After collecting a large amount of magnetic data,training samples are made by manual annotation.We divide geomagnetic signals into parking,leaving,and others categories.The training sample is the signal waveform after the geomagnetic terminal uploads the recovered data.we build a multi-layer 1D convolutional neural network to train magnetic signals.The results of the model show that the vehicle detection rate is 95.3% based on magnetic sensor signal acquisition and compression and convolution neural network signal classification algorithm.Our analysis shows that the proposed algorithm does not depend on the magnetic field base value,so it will not be affected by the drift of the base value.The proposed algorithm verifies that the magnetic signal waveform contains information that can be used for vehicle behavior classification.Besides,The proposed method is better than the traditional method in our experimental parking spaces.