Food Traceability System Based on 3D City Models and Deep Learning

Abstract

A 3D model-based food traceability system is proposed in this paper. It implements an information extraction method for processing video surveillance data. The first step of the proposed method is to build a 3D model of the target area. Based on the 3D models, cameras deployment in the surveillance system could be optimized with view coverage analysis. Then, we map the 2D views in video cameras into the coordinate system under the generated 3D models. Next, the deep learning based object identification method is selected to locate the interesting targets and their 3D coordinates are calculated based on the 3D model. Finally, multiple trajectories from different cameras are merged to create a complete traceability event for the target object. According to the experiment, the 3D models is useful to generate the unified traceability trajectories and the deep learning based method is efficient in extract the interesting objects from video surveillance system.