Improved ECO Algorithm Based on Residual Neural Network

Abstract Video target tracking is one of hot fields computer vision, and its application also very extensive. However, due to the complexity variability environment, which brings some challenges research tracking. ECO(Efficient Convolution Operators) Algorithm proposed based on convolutional neural network in three aspects. Firstly, residual ResNet50 adopted instead extract appearance features target, deeper applied obtain more abundant semantic information, so as improve effect algorithm.Secondly, sample space classification strategy improved. Different weights are assigned shallow feature deep feature, make play a important role Finally, method scale estimation improved that better bounding boxes can be estimated with changing. Experimental results show distance accuracy success rate algorithm.