LibrarySystem_Vision

Library Seat Management System Vision Module for CSC3002 group project in CUHKSZ
Complete repository of the project can be viewed in https://github.com/Doris619619/Fianl_Library_System

Brief

Detect the seat usage status with video model and record the data in jsonl files for the module B (temporal sequence juding) to judge if any seat is anomaly occupied, and save the annotated / original frame snapshot for the module C (database) to save into the database (if annotation function enabled).

Workflow

1. Specify the process approach (sampling frequency, max processing frame numbers, stream process or not (ignore this term if images are to be processed))
2. Read in images / videos at specific path, auto-detect the file type
3. Process the data:

• Size parsing
• Inference with Yolov8n (pretrained)
• Data Recording
• NMS
• Annotation
• Seat belonging judging
• Foreground ratio calculation
• Seat occupancy judging (current frame only: PERSON, OBJECT_ONLY, FREE, UNKNOWN)
• Sanpshot saving Iterate the process until all sampled frames done.

Model Fine-tuning Effect

Enable to detect person and 9 types of objects