Читать книгу Handbook of Intelligent Computing and Optimization for Sustainable Development - Группа авторов - Страница 74

Aniruddha Srinivas Joshi*, Savyasachi Gupta, Goutham Kanahasabai and Earnest Paul Ijjina†

Department of Computer Science and Engineering, National Institute of Technology, Warangal, India

Abstract

Garment detection is a complex image processing task that has a multitude of applications in the industry such as retrieval of similar garments, Artificial Intelligence–powered fashion recommendation models, and automatic labeling of catalogs. Retailers of fashion stores can benefit from knowing vital information about the types of garments that customers are interested in and thus ensure a more profitable business model. In this chapter, a novel framework is proposed for the detection of garments of interest from the footage of a surveillance camera. The video frames are processed using the GMG background subtraction model to obtain relevant foreground information along with foreground masks. The Mask R-CNN object detection model is used to identify customers and multiple other image processing techniques are used to obtain the active garments in these frames. The detected customers are tracked and the OpenPose human pose estimation framework is utilized on them to obtain useful landmarks. The garments of interest are then determined based on the filtration of confidence scores calculated for each active garment. The framework was tested on a CCTV video dataset and was found to be effective despite facing arduous obstacles such as background noise and occlusions.

Keywords: Garment detection, pose estimation, object detection, customer analytics, deep learning, computer vision