SAFE AND SECURE PLATFORM DEVELOPMENT FOR TRESPASSING RECOGNITION USING CCTV AND VIDEO ANALYTICS AT KAI STATION

• Main Author: Puteri Haryono, Hollyana
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/66353
• Institution: Institut Teknologi Bandung

Increasing mobilization in metropolitan cities makes trains become one of the most preferred transportation options by citizen. Smart mobility becomes one of the aspect that is realized in smart city. Bandung is one the capital cities in Indonesia that is realizing smart city in there. Bandung Station is the railway station that becomes our focus in this final thesis because it is the main railway station in Bandung City. Safety and Security is the most significant aspect for train passengers while doing a train journey and while waiting at the station. Solution that has been done in this final thesis is utilize installed CCTVs at station area. These CCTVs will be used for video analytics to people at the station, especially gate trespassing movements using multi object tracking methods. This final thesis aims to add trespassing activity recognition feature to increase safety and security on safe and secure platform in KAI station concerning the exactness of the chosen algorithm, model computation requirement, and model performance from precision score. Comparison have been done between 10 combination of detection algorithms (YOLOv3, YOLOv4, and YOLOv5) and tracking algorithms (SORT and DeepSort). YOLOv5 and DeepSort is the chosen and used algorithm in this final thesis. This algorithm combination rated the most accurate with almost equal computation requirement as other combination need. After adding the safe and secure feature to VIANA platform, features succeeded running on the platform with 92% GPU utilization, 19% memory utilization, 8 FPS of frame rate, and 44% precision score.