Abnormal Behavior Detection and Safety Early Warning in Electric Vehicle Network Interaction Systems Using Deep Learning Algorithms

In recent years, electric vehicles have become a trend in the automobile industry. However, while people get the convenience, electric vehicles also bring many hidden dangers to traffic safety as a result of their many abnormal driving behaviors. To reduce the occurrence of road traffic accidents, the study utilizes multi-task convolutional neural network and ensemble of regression tree feature point localization algorithm of deep learning technology to analyze and process the facial state of the driver and issue a timely safety early warning. The test results indicated that the accuracy of image recognition using ensemble of regression tree algorithm was 9% higher than that of tracking detection algorithm and the recognition speed was 61.7% higher. While compared with the multi-feature grayscale fusion algorithm, its accuracy and recognition speed were also improved by 5% and 52.5%, respectively. In the three kinds of abnormal behavior detection experiments, namely, driver calling on the phone, fatigued driving, and vision not focusing on the front, the time consumption by all the modules was below 60 ms, and the average processing time was 32.5 ms, which verified the high efficiency of the system. Experiments reveal that the detection and warning function designed in the study can provide drivers with abnormal behavior reminder service in a more accurate and real-time manner, which is of great significance in reducing the incidence rate of traffic accidents.