Research On Target Detection And Tracking Of Mine Pedestrian

Underground coal mines have uneven illumination,low illuminance and large dust,and video imaging is mixed with noise.There will be occlusion and high false detection rate during video monitoring.To ensure the safety of underground personnel,the detection and tracking of moving targets based on video surveillance information Safe production is of great significance.The image pro-processing and detecting and tracking methods in coal mine tunnel are studied.In order to solve the problem of low illumination and blurred image in video of coal mine,the different frequency components information is obtained by improving Retinex algorithm which adopting guided filtering,low frequency lighting components are corrected using Gammma function,high frequency refection components are enhanced and de-noiseing through CLAHE algorithm.The reconstructed image strengthens the outline information in detail and improves the contrast ratio of image.In the detecting of miner of coal mine tunnel,Gaussian background modeling cannot suit for complex and changing scenarios,deep semantic features and shallow representation features of moving target are extracted to realize the fusion of output features information in different convolution layers by combining the optimized YOLOv2 network structure,which improve the detecting effect and robustness of small target and nearby feature.Because the traditional target tracking algorithm exist some weakness,such as slowly update speed of model,the large calculations,which cannot meet the real time application,an improved twin neural network with the fusion of different level convolutional feature is proposed,which could track the similar miners effectively and has high tracking accuracy under the condition of real time.The simulation results shows that the enhancement method based on fusion of Retinex and Clahe can enhance the video image in coal mine tunnel,the entropy,average gradient and edge strength are better than the traditional algorithm;The improved algorithm which useYOLOv2 and Gauss model can realize the detection of miner and remove false alarm targets effectively and reduce the false detection ratio.The improved Siamese network target tracking algorithm can effectively track the miners in real time.The above methods have great significance for image processing and miner positioning in coal mine tunnel with low illumination video monitoring.