关键词: 车位检测, 迷你卷积神经网络, 本地响应归一化, 掩码, 机器学习

Abstract: Abstract: Aiming at the increasingly severe parking problem, a method of parking space detection based on a modified convolution neural network is proposed. According to the characteristics of parking space that could be represented by two states, the mini convolutional neural network(MCNN) is proposed, which is improved from the traditional convolutional neural network structure. The result shows that the method can effectively reduce the training time by reducing the network parameters. The local response normalization layer is added to the network to enhance the correction of the brightness and use the small convolution kernel to obtain more image details. After the training is completed, the video frame is masked and cut by edge detection into a single parking map, followed by the network for identification. The experimental results show that the recognition rate is 3% ~ 8% higher than that of the traditional machine learning mode, and the required network parameters are only 1/1000 of the general convolution model, so it is convenient to transplant in the low-profile camera.

Key words: parking space detection, mini Convolutional neural network, local response normalization, mask, machine learing