We found a match
Your institution may have rights to this item. Sign in to continue.
- Title
Efficient Method of Road Anomaly Recognition Using Deep Learning Coupled with Data Augmentation Approach.
- Authors
Natha, Sarfaraz; Jokhio, Fareed Ahmed; Shafique, Muhammad; Ahmed, Naeem; Arain, Danish Munir
- Abstract
In recent years, Closed-Circuit Television (CCTV) cameras have played a vital role in the surveillance of public and private areas. The primary objective of surveillance cameras is to monitor human behavior and road situations. In the real-world situation, detecting, and recognizing abnormal activities poses significant challenges due to the densely crowded environment and the composite nature of transportation systems. These factors make it problematic to automatically identify anomalies that occur while traveling leading to emergencies and threatening human life and property. In response to these challenges, Road Anomaly Recognition Surveillance Systems (RARSS) have been deployed to monitor and detect abnormalities on streets, highways, and roads. This research specifically investigated conventional and human-centered road anomalies encompassing snatching, accidents, fighting, and car fires. The proposed Real-time Road Anomaly Recognition system is formulated as a classification task, involving the analysis and processing of real-time CCTV videos. The proposed study investigated the deep learning model of Convolutional Neural Network (CNN) coupled with the Data Augmentation (DA) approach to address the frame flickering problem in real-time videos. Furthermore, we introduced a benchmark real-world Road Anomaly/Outlier Dataset (ROD) containing roads, streets, and highways videos and images with different illumination situations that presented various road anomalies. Experimentation using different pre-trained CNN models i.e., VGG19, InceptionV3, ResNet50, MobileNetV2, and DenseNet201 coupled with a DA approach has been performed on the ROD dataset. The experimental results demonstrated that the InceptionV3 model performed best with the augmentation approach compared to other deep learning models. It achieved the best accuracy of 98.80% in recognizing road anomalies.
- Subjects
DATA augmentation; CONVOLUTIONAL neural networks; DEEP learning; PUBLIC spaces; CLOSED-circuit television; HUMAN behavior; TRAFFIC cameras
- Publication
Sir Syed University Research Journal of Engineering & Technology (SSURJET), 2024, Vol 14, Issue 1, p76
- ISSN
1997-0641
- Publication type
Article
- DOI
10.33317/ssurj.627