选择性搜索和多深度学习模型融合的目标跟踪

Multi-Clue Fusion Target Tracking Algorithm Based on Selective Search and Deep Learning

提出一种基于深度学习的多模型(卷积神经网络和卷积深信度网络)融合目标跟踪算法.该算法在提取候选粒子方面,使用选择性搜索和粒子滤波的方法.CVPR2013跟踪评价指标(50个视频序列、30个跟踪算法)验证了:该算法在跟踪中能有效地缓解目标物体由于遮挡、光照变化和尺度变化等因素造成的跟踪丢失情况的发生.

A multi-clue tracking algorithm（convolutional neural network and convolutional deep belief network）based on deep learning was proposed.The algorithm used selective search and particle filtering method in extracting candidate particles.CVPR2013 tracking benchmark（50video sequences,30 tracking algorithms）verifies：the algorithm can ease the loss of tracking due to the occlusion,the change of illumination and size etc.