Wearable Recognition Using Artificial Intelligence
Image processing is used to extract important data from an image through processing and by using that information, machines can provide the understanding, interpretation, and description of the scene. Classifier allows us to process the image and map it to the provided classes. Our classifier allows
2025-06-28 16:36:44 - Adil Khan
Wearable Recognition Using Artificial Intelligence
Project Area of Specialization Artificial IntelligenceProject SummaryImage processing is used to extract important data from an image through processing and by using that information, machines can provide the understanding, interpretation, and description of the scene. Classifier allows us to process the image and map it to the provided classes. Our classifier allows the user to input an image, then the classifier identifies the wearables from the image using object detection techniques and classifies that wearable in provided classes. Object detection and classification is an important domain of artificial intelligence. Advancement in technologies provides the facility of building solutions, to automate recurring processes and eliminate human labor. Classification of products is a wider branch of object detection and computer vision, building the solution for the fashion industry to improve the workflow and for commercial use in the application of this solution. The basic categorization of wearable and apparel is done manually with the help of an attribute set of products. Each of these products lies in a class e.g. T-shirts, shirts, lower, shorts, dress, shoes, shades, caps, bags, etc. and each of them has an attribute set e.g. color, texture, color pattern, etc. The basic purpose of building the solution is to automate the classification process of these wearables by using advanced technologies. These classifications will be done by the classes of wearables and their attribute set provided in the dataset. Virtual Wardrobe is a concept of building a wardrobe of wearables or clothes a person owns, most of the time people do not know what they have and what to wear? So, using this solution, there might be an application that can facilitate people in building their virtual wardrobe.
Project Objectives- The main goal is to build a solution for the fashion industry as the fashion industry is using new technologies. Our solution may become the basic building block for developing advanced solutions e.g. Recommender System, VR & AR wardrobe, E-commerce Application, etc. The system will classify the wearables from input images using digital image processing and machine learning. After training, the model can identify the objects from an image like pants, shoes, shirts, tie, etc.
- Using the basic functionality of the wearable identification and classification model, the main goal of our project is to build a real-world application named virtual wardrobe. Virtual Wardrobe is a concept of building a wardrobe of outfits a person owns, some applications are providing the functionality of virtual wardrobe but they use simple techniques like clicking the photos and storing it also provide the list of outfits and ask from user about the outfit he/she owns, these application lacks to provide the fully automated solution and ease to the user.
- Our solution provides the functionality of either picking the pictures of user from the gallery or the user will select the photos of himself/herself from the gallery to automatically build the wardrobe by segmenting and categorizing the clothes from the photos. The application provides the collection of clothes, accessories and outfits to the user to quickly decide what to wear.
- We are planning to use the advanced Technologies of artificial intelligence and image processing, the project starts from the dataset collection, interviews, and surveys then we will move towards the preprocessing of the data which includes the cleaning, dividing and categorizing the data. After preprocessing the actual implementation of project starts which contains the Advance Image processing libraries like OpenCV which we use to process our collected dataset, creating, training and evaluating our model. This step needs high computational power to train a large amount of data present in the form of large resolution images. This dataset contains almost 250,000 images with different pose and different wearable with different colors. So here Azure Kinect Developer Kit helps us to analyze and process the images taken by the AI powered camera, the Azure Kinect DK has a strong CPU, computation power, AI capabilities, and Camera (Depth Camera). The usage of Azure Kinect DK is to track the body skeleton and predict the pose of the person present in the picture. Also, the Depth camera present in the Azure Kinect DK helps us to predict the physical measurement of the outfit/wearable. Lastly the amount of information comes from the Azure Kinect DK must be processed so for this purpose the GPU comes in, the NVIDIA GTX 1050ti 4GB OC Edition which can process large amount of image data, and information comes from the Azure Kinect DK. Lastly, we build the application of this solution analyzing the real-world problem.
- The application will be a mobile application using a mobile application framework like Android Studio, React Native, Flutter etc. The flow of the application will be the user select the photos from the gallery, application will pass these photos to our pretrained model of categorizing the clothes into classes, then the application will segment the clothes apart from the body and save it or just store the photo according to its class but it will depend on the user choice.
- The fashion industry is using these advanced technologies to make improvements and come up with innovative ideas to facilitate the audience and customer base. Wearable classifiers in the fashion industry help to make quick decisions overtime. This project has a great scope of utilizing the system's strength in the following ways:
- Clothing Recommendations: Recommended the alternative wearable by processing the input image under the rules of fashion ethics.
- Content Filtering: Filter the inappropriate content on media after identifying the inappropriate clothes.
- Visual Search: Search and Recommendations based on the input image.
- Building Virtual Wardrobe: Build the virtual wardrobe by taking images, separating the garments and storing them into a database and suggest the combination of clothes owned by the user.
- Detecting Replicas: Detecting replicas after classifying the input image.
-
The final deliverable will be a model which can identify the class of the wearable present in the picture. The model can be used as a basic building block for the implementation of advance application as discussed in the Benefits of the Project Section, or any other application which will propose in future. This model will be a pretrained model by a large amount of data with approximately 250,000 high resolution images. The model will be trained with the help of high-performance GPU named NVIDIA GTX 1050ti 4GB OC Edition which has capability of process high resolution images.
-
An application will also be included in the final deliverable. This application will provide functionality of building a virtual wardrobe, Virtual Wardrobe is a concept of building a wardrobe of outfits a person owns, there are some applications providing the functionality of virtual wardrobe but they use simple techniques like clicking the photos and storing it also provide the list of outfits and ask from user about the outfit he/she owns, these application lacks to provide the fully automated solution and ease to the user. Our solution will automate the entire process of building the wardrobe. In this scenario Azure Kinect DK will help us, Azure Kinect DK has a depth camera which has a capability to take the physical measurements of the object (wearables in our case). Also, the body tracking functionality of Azure Kinect DK helps to predict the best pose of the person present in the picture. The Azure Kinect DK also has some AI capabilities with pretrained models and computational powers which will help us in the future extension of the project like Augmented Reality Try room etc.
| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 70000 | |||
| Azure Kinect Developer Kit | Equipment | 1 | 55000 | 55000 |
| NVIDIA GTX 1050TI 4GB OC Edition | Equipment | 1 | 15000 | 15000 |