Abstract:
The use of drone views helps to create a safe transportation system by providing various traffic information. This paper aims to identify changes in ground vehicle moveme...Show MoreMetadata
Abstract:
The use of drone views helps to create a safe transportation system by providing various traffic information. This paper aims to identify changes in ground vehicle movements, such as stopping, lane changing, and safety, by tracking vehicles on the road. The vehicle condition is determined in two ways. First, the collected drone images are refined, and the data is augmented using the mixup method to identify the state of the vehicle. Second, a proposed learning model, the Wide Area Feature Extraction (WAFE), and Deformable Residual Module (DRM) are used. WAFE generates features by extracting objects across a wide area. DRM utilizes a deformable convolutional layer to extract features, incorporating information from the previous layer to create a feature map with receptive field flexibility. The experimental results indicate an 88.6 % accuracy for the vehicle state classification, with the model containing a total of 1.27M parameters. This represents a significant improvement over DRN_C_26, with a decrease 95% in the total number of parameters and a difference of 1.06 % in accuracy.
Date of Conference: 04-06 April 2023
Date Added to IEEE Xplore: 09 June 2023
ISBN Information: