Driver behavior feature extraction and discrimination method based on edge calculation

Abstract

The invention discloses a driver behavior feature extraction and discrimination method based on edge calculation, and the method comprises the steps: collecting video data of a driver, converting the video data into image data, building a face data set and a behavior data set, training the face data and the behavior data through a deep learning algorithm, and carrying out the identity authentication of the driver through a face authentication model; obtaining a region of interest of target detection and time sequence information of a video by using the detection model, training a driver abnormal behavior recognition model ADM, and judging whether an abnormal behavior exists or not according to a detection result of a video stream and the ADM model. According to the method, the cross-domain mobilenetv2 model is provided, the reasoning speed is improved, the combination of the time dimension and the space dimension is provided through the new abnormal behavior discrimination model, the region of interest in the space dimension is firstly extracted to serve as feature input, the feature space of the detection model is shared, the parameter learning time is shortened, and the model recognition precision is improved.

Description

technical field [0001] The invention relates to a driving behavior discrimination method, in particular to a driver behavior feature extraction and discrimination method based on edge computing, which belongs to the technical field of artificial intelligence. Background technique [0002] With the development of my country's social economy and the improvement of people's living standards, as well as the improvement of domestic road infrastructure development, in the field of municipal sanitation, the types and quantities of commuter vehicles in the field of sanitation are increasing, including: road sweepers, garbage transfer vehicles, sprinklers Wait. At the same time, artificial intelligence technology has developed rapidly in recent years, and technological development needs to be driven by the implementation of practical application scenarios. Among them, the safe driving of sanitation motor vehicles is a representative direction of deep learning technology in machine vis...

AI Technical Summary

Problems solved by technology

The biggest disadvantage of using the Gaussian mixture model to preprocess images is that the calculation is time-consuming, and the background is too complex to highlight features, which will directly affect the classification effect

If it is deployed to a low-profile edge computing box, it is difficult to meet the effect of real-time early warning

[0004] The second solution, such as Chinese invention patent 2015106526501.5, provides a vehicle-mounted intelligent reminder method and system based on behavior recognition. The system uses a two-step method. First, it needs to search for different positions and different scales in the video image to generate multiple recognition candidates area, which consumes a lot of computing resources; secondly, use the pixel-based cascade decision tree classifier to judge whether the candidate area is a face, which is difficult to identify a face in a changeable or complex environment

Images