Automatic Recognition of Driving Events Based on Dashcam Videos
Jui-Chi Chen, Zhen-You Lian, Chung‐Lin Huang, Cheng‐Hung Chuang · 2020
The recognition of driving events from dashcam videos is an important task in the intelligent transportation field. In this study, we compute the feature values of the dashcam videos using the optical flow method. We also propose a seven-region segmentation method to separate regions of different nature in a video frame. Then we use the random forest classifier to classify the output feature vectors into five driving event categories. We have tried a variety of different feature combinations for experiment. According to the experimental results, the recognition accuracy is up to 88%, and the average accuracy is 83.6%.