Research On High Altitude Throwing Object Detection Based On Adaptive ViBe Algorithm

The development of computer vision technology has brought great convenience to the community security and other fields.With the increase of throwing objects from height cases,it is necessary to protect citizens’ rights not only from legal means,but also from technical means to provide support.It is of great practical importance to study a comprehensive algorithm that can detect and track objects thrown from a height and apply it to the community intelligent security system to monitor and warn the problem of objects thrown from a height occurring in the community.Throwing objects from a height involves real-time detection and tracking of dynamic targets.Although deep learning can automatically learn distinguishing features from data to achieve accurate target detection,the model complexity of deep learning is high,the data computation is large,and the efficiency of model deployment is a difficulty in practical application scenarios where real-time processing of images is required.The currently commonly used ViBe background difference method is prone to the problems of inaccurate foreground segmentation and untimely background model update under the influence of noise and complex environment.Therefore,based on the analysis and research of existing dynamic target detection and target tracking technologies,this paper proposes an adaptive target detection and tracking algorithm for high altitude throwing objects by using cameras erected outside buildings to capture video of high altitude throwing objects.This paper focuses on the following aspects.(1)The target detection part,in order to achieve real-time detection of target objects,is proposed to use the ViBe algorithm with simple and easy to implement algorithm for detection.Firstly,for the traditional ViBe algorithm using fixed threshold foreground segmentation part,we propose to optimize the honey badger algorithm by using the backward learning strategy and the Corsi variation factor,and then calculate the optimal segmentation threshold of the image by using the two-dimensional OTSU that incorporates the multi-strategy honey badger algorithm to adaptively divide the image into two parts:foreground and background.Secondly,to address the problem that the ViBe algorithm uses a fixed update factor for background model update,we propose a method to dynamically adjust the update factor value according to the motion acceleration of the target object to achieve the consistency between the background update rate and the object motion change;finally,we perform morphological operations on the obtained foreground image to solve the problem of voids and noise.(2)The target tracking part,using a combination of Kalman filter and Hungarian algorithm tracking method.First,the Kalman filter algorithm will predict the motion trajectory of the detected target,get a predicted position track,carry out the next frame of target detection to get the detection position detection,by calculating the intersection ratio between the predicted position and the detected position,carry out the association of the target,then match by the Hungarian algorithm,only after confirming that it belongs to the same target,then carry out the Kalman filter algorithm for The position is updated and the tracking process of the target is completed by cyclic execution.Then,parabolic fitting is needed to identify and track overhead throws,while track ID matching,moving distance setting and multi-frame confirmation are used to distinguish overhead throws from other moving targets,so as to achieve online tracking of overhead throws.(3)The solution design section analyzes the difficulties of overhead throwing cases and proposes a basic solution for overhead throwing detection and tracking,including video data acquisition,region of interest setting,de-jittering research and video data cloud storage.The video acquisition part requires elevation angle to set up the camera outside the staircase to achieve full coverage of the building facade;ROI region setting effectively removes outside trees and other external environmental interference to improve the efficiency of target detection and tracking;the de-jitter algorithm research effectively suppresses the interference caused by camera jitter.Finally,the overall functional module design of the system is completed,and the effect diagram of high-altitude throwing object detection and tracking is given.The algorithm of high altitude throwing object detection and tracking achieves timely detection and real-time tracking for high altitude throwing objects,providing favorable technical support for community security.In summary,this paper conducts an in-depth study on high-altitude throwing objects from data collection,algorithm research and scheme design,and proposes an adaptive detection and tracking scheme to realize the target detection,real-time tracking,trajectory drawing and positioning functions of high-altitude throwing objects.Throwing object detection not only can prevent the occurrence of falling object accidents,but also is an important part of urban safety management,which can improve the modernization level of urban management and ensure the safety and smooth flow of urban traffic.At the same time,the research results of this paper promote the application of computer vision technology in the field of community security to a certain extent to broaden.