Bus passenger volume estimation method based on machine learning

Abstract

The invention relates to a bus passenger volume estimation method based on machine learning, and belongs to the fields of machine learning and intelligent traffic. A camera is erected in a bus to obtain a certain number of video images as training samples; the samples are marked, and an ROI probability density distribution map of each sample is calculated; the samples and the corresponding ROI probability density distribution maps are send to a deep convolutinal neural network for training to generate a model; and a video image to be detected is obtained, the model generated by training is used to regress an ROI probability density distribution map of the video image to be detected, all pixels in the ROI probability density distribution map are accumulated, and an accumulated value approximates to the passenger volume in the image to be detected. According to the method, disposition is easy, most time is spent in the machine learning stage, and in practical use, the trained model can be used directly, and passenger volume estimation is very low in computational complexity and high in computation speed.

Description

technical field [0001] The invention relates to the fields of machine learning and intelligent transportation, in particular to a method for estimating bus passenger volume based on machine learning. Background technique [0002] With the acceleration of urbanization, traffic congestion is becoming more and more serious. Bus travel has the advantages of low carbon and environmental protection, large traffic volume, and less road occupation per capita. Therefore, how to provide comfortable and convenient bus services and improve the public's experience of taking public transport is of great significance to attracting citizens to travel by public transport. For passengers waiting for the bus, if they can obtain the current location of the vehicle and the specific number of passengers in the bus, they can better choose the appropriate number of trains and arrange their travel plans reasonably; for the bus dispatching center, if they can obtain According to the number of passen...

AI Technical Summary

Problems solved by technology

[0003] At present, there are a variety of technologies used to count the number of passengers in the bus, such as infrared devices, video recognition detection, pressure pedals and other methods. Infrared devices and video recognition detection are to install infrared devices or cameras at the bus door passages to count each stop. The number of people boarding and getting off the bus is used to indirectly calculate the number of people in the car. This type of statistical method will have errors at each station. After a period of time, the errors will accumulate, and the number of passengers in the car calculated in this way may be very different from the actual number of passengers; pressure The pedal method has disadvantages such as easy damage to the physical device, difficult maintenance, and inability to distinguish the direction of getting on and off the car.

Images