Research And Implementation Of Personnel Target Tracking Algorithm For Video Surveillance System In Coal Mine

There are many coal-mining enterprises in China,and the safe production in the coalmine field is a great significance to the development of society and economy.Video surveillance is a key measure to ensure the safe operation of underground coalmine.Realizing the tracking of underground personnel can reflect the information of the miners in the mine in real time,which is beneficial to the positioning of underground personnel and system warning under dangerous conditions.It is also a great significance to the safey of underground miners and equipment and the automation of monitoring systems.In this paper,for the monitoring scene of coalmine underground,in the representative scenes of the mine,the Camshift algorithm combined with color and texture features and the YOLOv3 network framework in deep learning are used to realize real-time detection and tracking of the miner.The work of this paper mainly includes the following aspects:(1)Tracking of the miner at underground roadways and coal conveyor belts.There are some noise interference and extremely uneven illumination in the downhole images.Therefore,the image is preprocessed to improve the image quality,especially the quality of the initial frame.In view of the characteristics of the miner here,the tracking area with distinctive color and texture features,that is,the area above the shoulder of the miner,the H component of the HSV color space is selected for the color feature,and the improved equivalent LBP texture is selected for the texture feature.Referring to the fusion of two characteristics of probability density function in Mean shift algorithm,and we use Camshift algorithm to track miner.Experiments have shown that miners can be tracked accurately and in real time in the presence of dimly lit light and headlight interference.(2)Identification and tracking of miner at underground substation.Downhole substation is important for equipment,and the safety of equipment and miner are particularly important.In order to achieve more accurate detection and tracking of the miner,this paper chose the YOLOv3 deep learning network framework.The network structure reduced on the original feature extraction network,and modifying the network parameters.The miner image data of the substation are preprocessed and retraining the miner data set of the downhole substation.Experiments show that the modified training model converges more quickly and smoothly.Under the condition that the miner has multiple attitudes,headlight interference and becomes a small target,under the premise of setting the confidence score of 0.25,there is no frame loss phenomenon for the miner in the video,and it meets the needs of real-time monitoring.(3)Analysis of miner's behaviors in substation.After tracking the miner in the substation,simply analyzing the follow-up behaviors of the miner.First,cross-border detection,using the positional relationship between the center point of the miner-predicting box and the safety line to determine whether there is a cross-border situation.Second,detention detection,statistics and analysis of substation miner's inspection time,through the miners in the delimited area of time,determine the miners' detention.It can improve the safety awareness of miners and ensure the safe and stable operation of substations.The detection and tracking of underground the miner has improved the automation degree of the mine monitoring system,which is of great significance for the safe operation of underground personnel and the safe operation of equipment.