A Dynamic Recommendation System of Time-Varying and Geotagged Images from Surveillance Cameras

抄録

Cherry-blossom viewing and leaf peeping are very popular activities. They can not only promote the mood benefits, but also create great economic profits to nearby areas. However, even in this information explosion era, most people still rely on news, comments/recommendations on SNS when they are searching for a good viewing spot. At the same time, with the advances of Information and Communication Technologies, the surveillance cameras have covered nearly everywhere in the traffic networks of smart cities now. The above mentioned life related needs, infrastructure changes, and technologies advancement motivated this research. In this project, we proposed a dynamic recommendation system which combined AI technology and real-time cameras images. To achieve this purpose, we developed an image database by using the image resources of Google Street View. And then we analyzed the color changes of the leaves when it has variation. After the preprocessing, we semantically segment the shapes of trees by adopting our pre-trained deep learning model. Then we combined the real time results with other botanic related information changing in one year for verification. The current experimental result shown that our system can accurately recognize the latest status of street trees for making possible predictions. Based on these dynamic predictions, users can make better itinerary planning for Hanami.