Automatic detection of triples rider violations using an optimized mask R-CNN framework

Road safety is a critical concern in urban and rural areas, with triple riding on two wheelers posing significant risks. This practice not only compromises stability, control and visibility but also violates traffic regulations, leading to severe accidents and road congestion. To address these issues, a novel study was conducted to detect triple rider violations in real-time environments. This study developed a live environmental detection model that categorizes data into four classes: double riders, triple riders, double riders with motorcycles and triple riders with motorcycles. The model employs a modified Mask R-CNN architecture for the classification and segmentation of triple riders, achieving high accuracy across various climatic conditions using extensive training datasets. To enhance the performance, the model incorporates two deep neural training networks: Dense Network-121 layers (DenseNet-121) and Residual Netwrok-50 layers (ResNet-50). The effectiveness of the system was evaluated using multiple metrics, including the Intersection over Union (IOU) matching score, Accuracy (A), Precision (P) and F1 Score. The experimental results demonstrate the model's exceptional performance, with triple rider detection accuracy reaching 97.93%, doubles/triples rider classification at 97.41%, triple rider segmentation at 98.22% after 10,000 training steps. This Integration of AI-based triples violation detection,contributing to organized traffic and transportation.