YOLO-based enhancement of public safety on roads and transportation in Bangladesh

Abstract

In order to upgrade the efficiency level of multiple tracking like face, actions, characters, a deep learning method is introduced to reduce the accidents occurred in roads for carelessness and also to capture the criminals in Bangladesh. This paper presents a faster processing multiple detection method with the best possible outcome under the framework of YOLOv2 algorithm in the event of car accident, crossing foot over bridge and using the zebra crossing in Bangladesh. Different layers were added to the YOLOv2 algorithm to pass the information in various convolutional layers to detect multiple objects with actions. In this paper YOLOv2 algorithm under DarkFlow framework is used to achieve higher ratio of confidence value as the max convolutional layers reorganize the feature map so that other layers feature map can be matched with the bottom layers to achieve the expected output of the indicated events. By removing the noise from the unrelated area, the detections of the training video and test video adopt quite parallel confidence ratio