The use of skateboards for transportation in pedestrian dense areas is becoming more prevalent. Increased use of skateboarding raises the probability of pedestrian-skateboarder collisions and near-collision events. Skateboarders and pedestrians can face significant injury when involved in a collision. New approaches are needed to measure the frequency and predict the potential for collision events, in real-time, as traffic conditions change due to construction or land usage. Surrogate Safety Measures can be computed to assess hazard conditions on roads and sidewalks using deep learning object detection and classification models. We developed a new dataset consisting of over ten thousand images with nearly thirty thousand bounding box annotations of pedestrians and skateboarders at eighteen different camera perspectives. We trained the Faster R-CNN and SSD models with our dataset. While both models were found to correctly classify pedestrians and skateboarders with images containing both classes, the Faster R-CNN model performed with greater accuracy than the SSD model. However, the SSD model was shown to classify at a higher video frame rate which makes SSD a candidate for edge-based detection and classification and lays the ground work for automating the calculations for Surrogate Safety Measures between skateboarders and pedestrians.
Software And Hardware
• Hardware: Processor: i3 ,i5 RAM: 4GB Hard disk: 16 GB • Software: operating System : Windws2000/XP/7/8/10 Anaconda,jupyter,spyder,flask Frontend :-python Backend:- MYSQL