Detecting Helmet of Bike Riders in Outdoor Video Sequences for Road Traffic Accidental Avoidance

Abstract

In metro cities of all over the world, the growing number personal vehicles and fast life style of the people frequently meet very serious accidents. Due to deeply regretted reports from the loss of manpower and economy, accidental avoidance becomes a hot challenging research topic. In this paper, we consider specifically the accidents that happen due to bike rider’s involvement. Focusing on detecting helmet test, we proposed a computer vision based model that exploits HOOG descriptor with RBF kernel based SVM classification. Our experiments have two tier classifications, first is between bike riders and non-bike rider’s detection and second is to determine whether the bike riders in the first phases wearing a helmet or not. The initial phase uses video surveillance for detecting the bike riders by using background modeling and bounding based object segmentation. The performance comparison of our model on three widely used kernels ensures the validation of the satisfactory results. We achieved helmet detection accuracy with radial basis kernel 96.67%. Our model can detect any type of helmets in the outdoor video sequences and help security and safety aspects of bike riders.