Convolutional Neural Network Customization for Parking Occupancy Detection

A lot of Research has been developed to improve the reliability of smart parking systems. The utilization of computer vision is more beneficial than using sensors for the detection of parking spaces on smart parking systems. One smart camera can monitor multiple parking spaces according to camera sensing. The use of cameras is more efficient than using sensors, because sensors require expensive installation and maintenance costs. The accuracy and computational time are challenges that must be resolved by applying computer vision to classify the parking spaces. We perform several comparisons of computer vision classifications by utilizing the pre-trained Convolutional Neural Network (CNN). The mAlexnet customization is called by CmAlexnet to improve accuracy in classification. Parking space classification for CNRPark Camera A and B dataset was done. GoogleNet, Alexnet, VGGNet, mAlexnet, and CmAlexnet had varied results. From the testing done states CmAlexnet is better in the accuracy of parking space classification. The average accuracy of CmAlexnet outperforms all pre-trained with almost the same training time as mAlexnet. The test result stated that the performance of mAlexnet can be improved.