Author:
MEHMET ERKİN YÜCEL
Year:
2022
Abstract:
Smart retail stores are becoming a fact of our lives. Several computer vision and sensor-based systems work together to achieve such a complex and automated operation. Besides, the retail sector already has several open and challenging problems which can be solved with the help of pattern recognition and computer vision methods on embedded systems. Two important problems to be tackled are the stock-out (out-of-stock) and planogram compliance control. In this thesis, we propose novel methods to solve these problems. First, we frame the shelf control operation as change detection for the stock-out problem. Due to the constraints in retail stores, systems with ultra-low and low-power microprocessors with embedded cameras are formed. The change detection methods in the literature are adapted and used to work on microprocessors. We tested all the proposed change detection methods on our custom dataset and summarized the key findings regarding price, detection performance, and battery life. Second, we proposed a method based on object detection, planogram compliance control, and focused and iterative search steps for the planogram compliance control problem. Local feature extraction and implicit shape model formation form the object detection step. The planogram compliance control step is formed by sequence alignment via the modified Needleman-Wunsch algorithm. The focused and iterative search step aims to improve the performance of the object detection and planogram compliance control steps. We tested all three steps on two different datasets and summarized the key findings as well as the strengths and weaknesses of the proposed method. Additionally, a hardware platform with a wireless module, photovoltaic cell, rechargeable battery, power management module, and PIR sensor is proposed such that the proposed shelf control systems can work wirelessly in retail stores. The implementation details of the system and test results are provided both from a software and hardware perspective.