Abstract:
This paper deals with real time segmentation of traffic images using a Mask R-CNN model. The aim is to improve the performance of real time image segmentation, so that it...Show MoreMetadata
Abstract:
This paper deals with real time segmentation of traffic images using a Mask R-CNN model. The aim is to improve the performance of real time image segmentation, so that it can be effective even with noisy images captured by traffic cameras. The approach developed here comprises of image pre-processing, object detection followed by segmentation. Mask R-CNN model not only segments the image, but also surrounds the image with bounding boxes and assigns class names to the individual objects e.g. Car, Truck, Bus, Bicycle, Person etc. The model is trained with the annotated MS COCO Training dataset. To improve the performance of Mask R-CNN over noisy images, here pre-processing algorithms like Non Local Means (NLM) filter denoising and Median filter denoising are used. The testing is carried out on a subset of MS COCO Test dataset which comprises of only traffic images. The improved performance is demonstrated using parameters: increased correct object detections and corresponding confidence value, reduced incorrect object detections and corresponding confidence value, and an overall enhanced segment mask area accuracy.
Published in: 2018 9th International Conference on Computing, Communication and Networking Technologies (ICCCNT)
Date of Conference: 10-12 July 2018
Date Added to IEEE Xplore: 18 October 2018
ISBN Information: