Enhanced Spatial Pyramid Pooling And Intersection Over Union In Yolov4 For Real-time Grocery Recognition System

The ability to recognize a grocery on the shelf of a retail store is an ordinary human skill. Automatic detection
of grocery on the shelf of retail store provides enhanced value-added to consumer experience, commercial
benefits to retailers and efficient monitoring to domestic enforcement ministry. Compared to machine visionbased object recognition system, automatic detection of retail grocery in a store setting has lesser number of
successful attempts. In this paper, we present an enhanced YOLOv4 for grocery detection and recognition.
We enhanced through spatial pyramid pooling (SPP) and Intersection over union (IOU) components of
YOLOv4 to be more accurate in making recognition and faster in the process. We carried an experiment
using modified YOLOv4 algorithm to work with our new customized annotated dataset consist on 12000
images with 13 classes. The experiment result shows satisfactory detection compare to other similar works
with mAP of 79.39, IoU threshold of 50%, accuracy of 82.83% and real time performance of 61 frames per
second