SeeU Track

In this porject, We develop AI model for AI CCTV system which is called SeeU track. For this project, the system is mainly focus on the re-identification task for vehicle and person. For Vehicle re-identification, we employed object detection model for detecting vehicle and its license plate, classification model to recognize the vehicle's brand, model, and color, and text detection and text recognition model (OCR) to read the license plate. Besides, for person re-identification, we applied object detection model to detect person and person attribute recognition model to recognize the person's attribute in terms of gender, age, type and color of cloth, etc.

We exploit Object Detection model to detect person, vehicle, and its license plate to detect the target accurately. To select the most decent model in the desired environment, we searched and tested the open source model, and decide to use YOLOv11 model.

We use Object Classification model to extract the MMC (maker, model, color) Information of the vehicle. For this task, many open source models are tested and YOLOv11 model is chosen since it has a higher accuracy than others.

OCR model is used for this task. OCR model is consist of text detection and recognition. We tested and evaluated models and dbnet and svtr models are chosen.

Person Attribute Recongition is the task to extract the information of the person including gender, age, type and color of cloth, accessory, etc. To develop the model, we tried different architectures to extract the feature information such as Resnet, vision transformers (ex. swin transformer), and etc. After many test, we decide to employ vision language model for extraction and transformer for classification task.