Research On Foreign Object Detection On Urban River Surface Based On Machine Vision

With the acceleration of China's urbanization process,the situation of water resources protection is becoming increasingly severe.In particular,problems such as water environment pollution,aquatic ecological damage,and water shortages have become increasingly prominent,which has become a bottleneck restricting the development of society.Foreign matter invasion due to various reasons such as domestic garbage disposal,factory illegal discharge,surface runoff scouring,etc.will cause serious damage to the self-purification ability of the river water ecosystem.It also causes eutrophication of the water body,imbalance and destruction of the water ecosystem,and deterioration of river water quality,which seriously affects the surrounding living environment.In order to promote the protection of the river and lake system and the overall improvement of the aquatic ecological environment,China has comprehensively implemented the "river chief system",but it still relies on manual inspection at the river scene and discover problems through visual observation,which has large workload and low efficiency.It is urgent to use machine vision and intelligent analysis technology to realize the automatic detection of foreign objects on the river.However,due to the particularity of riverine environment,such as different river structure,complex water surface appearance,and changing natural conditions,etc.,there are still many difficulties to be solve by using existing machine vision methods to detect foreign objects in the river.Therefore,this paper has carried out research on the method of foreign objects detection on the river based on machine vision.The main research work and innovations are as follows:(1)A general water body segmentation algorithm was proposed based on the study of the imaging mechanism of water pixels,which aim to deal with the different appearances of rivers and complex background noise in different outdoor scenes.The algorithm recognizes water region based on the Multi-Block Local Binary Pattern(MB-LBP)and the variance of the hue value,according to the law that the pixel intensity changes with the distance and the stability of the hue value of pixels under light and shadow.This algorithm can segment river scenes with different structures and appearances without prior training.Compared with traditional methods,the algorithm has higher Pixel Accuracy(PA)and Mean Intersection over Union(MIoU).The region of water body segmented by this algorithm is regarded as the Region of Interests(RoI)of the subsequent detection algorithm,which is an important preprocessing step in this paper.(2)To solve the problem of high false detection rate of traditional object detection algorithms under the complex water surface background,the image sequences in the video is used to construct a Gaussian Background Model in the HSI color space,thereby enhancing the robustness to light and shadow changes.A detection algorithm for foreign objects on river surface utilizing multi-frame and background differencing is proposed.This method combines differencing results between successive multiple frames and the background model to reduce the probability of false detection caused by random noise of the river surface.A parallel algorithm framework is designed,where the background updating and object detecting are executed in parallel,which improves the structure of the algorithm.The experiment proves that the designed algorithm can more accurately detect the foreign objects under the complex background of river surface.(3)To improve the poor tracking of objects under the complex riverine background and the inability to update the target frame size,the improvement for Kernel Correlation Filter(KCF)algorithm is proposed.Based on the KCF algorithm,a multi-feature fusion decision-making mode is designed.The HOG features extracted from both HSI and RGB color space is combined to enhance the tracking ability of the KCF algorithm under the riverine background.A method based on multi-scale response analysis is also designed to make the target frame adaptively update during the tracking process.The experiments show that the improved KCF algorithm has better performance in river scenes,and improves the ability to track foreign objects under complex riverine background.(4)The online monitoring system for floating foreign objects of urban rivers is designed and developed.Based on cloud computing and IoT technology,the system is built utilizing Spring MVC framework,and its key functions such as online processing of data and visual display of results are implementedIn conclusion,research on the detection method of foreign objects in urban river scenes based on machine vision is carried out in this paper.After experimental verification,the method designed in this paper can effectively detect the foreign objects in urban river scenes.This research is of great significance to help improve river environment and reduce the cost of labor,and also provides a solution for the field of object detection and tracking under complex backgrounds.