| Computer vision technology has become an essential element to assist urban management,greatly reducing the cost of target information detection,and has high application value in pedestrian height detection,vehicle speed detection and other fields.Compared with physical measurement methods,information measurement with the blessing of computer vision technology has the characteristics of non-inductive,versatile and efficient.However,computer vision technology generally needs to use special equipment,such as binocular cameras,cameras with sensors,etc.,and replacing the monocular cameras that have been deployed in large numbers in cities with these professional equipment will consume a lot of costs,and also bring limitations to the research of computer vision measurement algorithms.Based on the above practical requirements,this paper proposes an out-of-the-box monocular visual measurement algorithm,which can complete the task of detecting the height and distance of moving targets without adding other hardware devices and configuring only some parameters required by the algorithm.This algorithm is suitable for applications where it is necessary to retrieve fast-moving targets and accurately measure their height and distance.In this paper,the open-source dataset Pascal Voc is used as the dataset,and pedestrians are more common in cities as the detection target,and the dataset is customized by combining manual labeling and model self-training,and a variety of data enhancement methods are combined to improve the data quality.Yolov4 is selected for the algorithm of detecting the target,and the processed dataset is trained as an algorithm model that meets the requirements of this paper.Based on the principle of object detection,the model cannot detect the accurate location coordinates of the target,and the resulting error will lead to the deviation of the measurement results.Therefore,this paper proposes a chunked adaptive background difference algorithm,which can remove the influence of complex background.Combined with the global contour detection method,the true contour position of the target in the frame is obtained,so as to calibrate the position information of the target.In terms of measurement algorithm,this paper analyzes the basic process of monocular imaging based on the principle of geometric perspective,and proposes a formula for measuring height and distance under monocular visual conditions based on the principle of geometric perspective,combined with relevant prior parameter information.By substituting the calibrated target coordinates into the formula,the height and distance information of the target can be calculated.In this paper,through pedestrian detection experiments in real environments,it is verified that the accuracy of the target distance measurement is above 97% and the height accuracy is above 95% within the range of 24 m.Experimental results show that this method can be applied to detect the height and distance information of moving targets. |
