With improvements in edge computing chip performance and optimization of deep learning models, the accuracy of video-recognition parking solutions has reached over 95%.