A driver abnormal posture monitoring method based on reinforcement learning

Abstract

The present invention claims to protect a method for monitoring abnormal posture of drivers based on reinforcement learning. The facial key points are used to construct facial features through cubic spline interpolation for driver identification. The key frame extraction method is based on reinforcement learning, obtains the corresponding reward through the posture detection model, and updates the action value function according to the reward and each action of the posture until a stable key frame extraction strategy suitable for each driver is obtained. Based on the temporal and spatial changes of the driver's dynamic behavior, the machine learning algorithm is used to train the attitude detection model. Combined with the theory of protection motivation, a safety early warning mechanism with guiding nature is established. The invention increases the real-time performance and accuracy of detection, and enhances the reliability of safety warning.

Description

technical field [0001] The invention belongs to the technical field of safe driving detection, and in particular relates to a method for monitoring abnormal posture of drivers based on reinforcement learning. Background technique [0002] As a means of transportation, the car has become an indispensable necessities of life. With the rapid growth of the number of private cars and commercial vehicles, it is of great significance to detect and warn the driver's driving behavior for reducing traffic accidents. [0003] Driving behaviors are divided into two categories: the state of the driver inside the vehicle and the state of the vehicle outside the vehicle. The research on driver behavior detection is divided into two categories: the traditional sensor-based wearable detection method, which not only causes interference to the driver, but also has high equipment cost; the mainstream detection method is the detection method based on monocular camera , the method obtains a lar...

AI Technical Summary

Problems solved by technology

[0004] However, the driver behavior research based on the monocular camera still needs to solve the following problems: for some reasons, the images captured by the on-board camera are not clear, which will cause some changes in the extracted driver's physical characteristics, and the traditional detection results will be confused. ; In most cases, when training the model, an image or even an image sequence is used as input, which will require a lot of time and resources to train the detection model. The amount of video data is too large to achieve real-time effects; most of the current detection models are trained based on big data, often considering the general characteristics of ordinary drivers' driving behaviors, while ignoring the behavior characteristics of different drivers

This method needs to collect a large amount of data in defined scenarios for training, and the emerging dangerous driving behavior cannot be recognized or the recognition accuracy is low; if all scenarios are defined and details are detected, the range of use of the model is limited; and based on The model training method of universal features does not consider the action characteristics of different drivers, which will cause deviations in detection or untimely detection results

[0006] At present, there is no method to detect the driver's normal driving behavior, bad driving behavior (calling, drinking water, looking at the mobile phone, etc.) and dangerous driving behavior (fainting, etc.) based on reinforcement learning and key point information of human bones. problem, this method proposes a driver behavior detection method based on reinforcement learning to correctly identify and warn potential dangerous driving behaviors

Images