Fast Classifying Non-helmeted Motorcyclists by Using Convolutional Neural Networks

Abstract

According to statistics from ThaiRoads foundation, fifty percent of motorcyclists in Thailand ignore helmet while riding. This reckless behavior may lead to a severe injury when they encounter an accident. Most existing non-helmeted detection techniques analyze all moving objects in the video frame without ignoring unrelated pieces. They need to classify a large number of moving objects in the buffer, resulting in a delay in the execution time. We propose the fast system to accurately detect non- helmeted motorcyclists from the surveillance cameras by using the object filtering technique along with two convolutional neural networks (CNNs). The first CNN identifies motorcycles among filtered moving objects while the second CNN detects non-helmeted motorcyclists. The experimental results on the video stream dataset captured from surveillance cameras at King Mongkut’s Institute of Technology Ladkrabang indicate that the proposed system not only improves an execution time but also produces a higher classification accuracy, comparing among other algorithms.