Train driver action real-time recognition system and method based on vehicle-mounted GPU acceleration

Abstract

The invention provides a train driver action real-time recognition system and method based on vehicle-mounted GPU acceleration, a real-time monitoring module is connected with an object detection module, the object detection module is connected with an action recognition module, and the action recognition module is connected with a monitoring and early warning module, a storage module and an information transmission module; the recognition method comprises the steps that the real-time monitoring module collects a driver monitoring picture; the object detection module identifies all body partsof a driver and important marked objects in a cab; the action recognition module performs action matching to obtain a recognition result; and the subsequent processing module processes the identification result. The invention provides a train driver action real-time recognition system and method based on vehicle-mounted GPU acceleration. Aiming at the characteristics of the vehicle-mounted GPU, the deep learning model is optimized and improved, and on the premise of ensuring the recognition accuracy, the driver action recognition speed is greatly improved, so that the driver action recognitionspeed can be operated in real time in an embedded environment.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a system and method for real-time recognition of train driver actions based on vehicle-mounted GPU acceleration. Background technique [0002] Monitoring the behavior of the train driver during the train running is of great significance to ensure the safety of the train. With the rapid development of deep learning and computer vision, many excellent human action recognition solutions have emerged. However, these solutions often require heavy calculations, have extremely high requirements on hardware, and are difficult to run in real time. For the specific problem of train driver action recognition, CMU's OpenPose human body key point detection system has too high requirements for hardware. It can only reach a speed of about 5fps on the GTX 1080, which cannot meet real-time requirements; Various deep learning action recognition systems are difficult to deploy on trains ...

AI Technical Summary

Problems solved by technology

However, these solutions often require heavy calculations, have extremely high requirements on hardware, and are difficult to run in real time.

For the specific problem of train driver action recognition, CMU's OpenPose human body key point detection system has too high requirements for hardware. It can only reach a speed of about 5fps on the GTX 1080, which cannot meet real-time requirements; Various deep learning action recognition systems are difficult to deploy on trains due to high computational load and large models, resulting in high hardware requirements and difficult real-time guarantees.

Images