Triple Riding and No-Helmet Detection

Two-wheelers or motorcycles are the most used and popular mode of transport all over the world. But, there is a significant risk involved if a motorcycle rider does not follow all the traffic rules. If a rider doesn’t wear helmet, the risk of losing life during an accident is significantly higher. The problem with the existing traffic surveillance system is that it is not so efficient as it involves intervention of humans, whose efficiency may not be ideal, and this has led to increasing road accident rates and fatality rates due to road accidents. Hence, this project aims at solving the problem by automating the task of detecting the riders without helmet and the riders who are triple riding. A traffic video as an input is given to the system. The system recognizes the motorcycles in the traffic scene by using the object detection model which is built using YOLOv3. If the system detects a two-wheeler, then another YOLOv3 object detection model is used for helmet detection. For triple riding detection, the riders are recognized with the YOLOv3 model for person detection and the number of detected riders is counted. If the number of motorcycle riders exceeds two, then it indicates triple riding, and the image of the rule violating motorcyclists is captured, saved and an alarm is buzzed. The system was tested for accuracy on a total of 100 videos. The accuracy of motorcycle detection was 99%, for helmet detection was 92% and no-helmet detection was 97%. The accuracy for one rider is 99%, for two riders is 96% and for three riders is 95%.