研究目的
To robustly estimate product amount on store shelves from a surveillance camera for improving on-shelf availability in retail stores.
研究成果
The proposed method accurately estimates product amount on shelves with a success rate of 89.6% within an error margin of one product, enabling improved on-shelf availability and business profit in retail stores. Future work could involve error reset mechanisms and better region segmentation.
研究不足
The method may fail due to classification errors that accumulate over time, and issues with detecting multiple change regions as one region, which could be improved with image segmentation or regular error resetting. Low resolution and frame rate of videos may also pose challenges.
1：Experimental Design and Method Selection:
The method involves detecting change regions using background subtraction and moving object removal, classifying these regions with a convolutional neural network (CNN) based on CaffeNet, and updating shelf condition to compute product amount.
2：Sample Selection and Data Sources:
Two videos captured in a real store with durations of 185 and 95 minutes, resolution of 480x270 pixels, and frame rate of 1 fps. Training samples were extracted from other videos in the same store.
3：List of Experimental Equipment and Materials:
A fixed surveillance camera attached on the ceiling, videos, and a computer for processing.
4：Experimental Procedures and Operational Workflow:
Change regions are detected, tracked using the Hungarian method, classified into four classes (e.g., product taken, replenished), and shelf condition is updated to compute on-shelf availability every minute.
5：Data Analysis Methods:
Success rate is computed based on error margin between estimated and ground truth on-shelf availability, using statistical evaluation.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
