Abnormal Activity Detection and Classification of Bus Passengers with In-Vehicle Image Sensing.

ROHINI A , Dr. ARAVINDA T.V , Dr. KRISHNAREDDY K.R , Prof. RAMESH B.E , NAYANA R

YOLOv8, VGG16, falling, sitting, Video surveillance

Traditional video surveillance systems in public transportation rely heavily on manual monitoring, which is time-consuming, error-prone, and inefficient in detecting anomalous passenger behavior in real time. This project proposes an intelligent, automated system for detecting passenger actions and identifying anomalies within a bus environment using deep learning techniques. The proposed system integrates YOLOv8, a real-time object detection algorithm, to identify and track passengers from live video streams. Once passengers are detected, the system extracts individual regions of interest and classifies their actions using a fine-tuned VGG16 convolutional neural network. The model is trained to distinguish between normal behaviors (e.g., sitting, standing) and anomalous actions (e.g., falling, squanting), enabling timely alerts for safety interventions. This dual-stage approach—combining spatial detection with semantic classification—ensures accurate and robust identification of potentially hazardous situations. The system supports real-time processing and can be enhanced with alert mechanisms such as visual warnings, sound alarms, or event logging. By automating the detection of unsafe or unusual behavior, the system significantly improves passenger safety, reduces the burden on human operators, and contributes to smarter public transportation surveillance systems.

Volume 10 Issue 8, August-2025

Pages : a20-a28