Gully Sensors
As flooding becomes an increasingly larger issues in many cities, solutions that can help to alleviate the damage caused by flooding are in high demand. By using video and image based sensors, Gully sensors detect when blockages in gullies occur so that they may be unblocked and flooding damages avoided.Drainage gullies can become blocked very quickly when materials like sludge or gravels are carried onto the road during heavy rain or flooding. This can also occur when there is a heavy fall of leaves. Monitoring drainage gullies improves the effectiveness on understanding the impact of gullies blockage in flood risk management schemes, enabling citizen groups to collect information and report incidents by devising innovative ways they can gather data, emergency works plan in emergency response to flooding caused by blocked gullies as well as obtaining new and real-time data on the status of each gully.
The level of blockage of gullies and drainage is one of the many important parameters for local authorities and city councils to monitor in real-time. Artificial intelligence (AI) tools are being considered on categorising the drainage gullies blockages images. Images are classified into multiple class labels using deep learning technique.
An image classifier is being developed, led by the University of Bradford, for Video/Image based sensors (i.e. Gully Sensors). DCNN (Deep Convolutional Neural Networks) is applied with limited training image data set to classify drain blockage incidents. The training images are collected from Google images/ YouTube videos and other sources. A scalable approach is used to classify the images by exploring the context of images by applying image segmentation. Image segmentation localise the area of interest in the images before the images are utilised for the training and validation phases. The preliminary results showed that the image classification using image segmentation at data pre-processing stage has higher accuracy in comparison of the classification accuracy of the image classification without image segmentation.
For more information, please contact:
Dr Dhaval Thakker: D.Thakker@bradford.ac.uk
