Road Accident Detection using MobileNetv2 with Gradient Boosting Algorithm

An increased number of vehicles on the roads increases the risk of accidents and eventually resulted huge loss of lives and property. Accident detection utilizing computer vision in much consideration as of late. The existing methods for accident detection gave inaccurate detection due to the errors in sensors and cameras may not capture all angles or all areas. Therefore, this paper proposes MobileNetV2-Gradient Boosting (MNGB) for detection of road accidents which effectively gave accurate detections. Firstly, all videos are split up into smaller shots according to scene changes and then key frames are extracted from each shot based on consecutive frames. This images are augmented by using Kalman filter which reduced the noise and artifacts in images, accurately tracks the objects across frames. Then, features are extracted by using Multi-Scale Network (MSNet) where it extracts the features at various scales by capturing both local and global information. After that, these features were classified by using proposed MNGB which enables the real time processing and reduces the computational resources and gave accurate detection results. The proposed MNGB for road accident detection achieved better results in terms of accuracy $95 \%$, precision ${9 8 \%}$ and F1-Score of ${9 6 \%}$ when compared with existing YOLOv3.