Target motion tracking method based on brain-inspired memory network

• Main Authors: Yong Wang, Fengli Shen, Jiapeng Zhang, Linwei Zhu, Jiamei Liang, Xinghua He, Rui Qi, Yanchun Zhao, Huanlong Zhang
• Format: Patent
• Language: English
• Published: 16.10.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

Method of target motion tracking based on a brain-inspired memory network, comprising: using a visual geometry group (VGG16) backbone network (feature extraction network) to extract features from an initial image and store the target in the brain-inspired long-term memory network; extracting features from a subsequent image using the backbone network; inputting the subsequent frame's feature information into the brain-inspired long-term memory network and using an attention-driven module (channel and spatial attention) in combination with previous memory content to predict a target position; if the confidence of the tracking results is lower than a threshold, processing the next frame; and if the confidence of the tracking results is greater than a confidence threshold, writing an extracted feature of the target into a memory internal storage, and updating the brain-inspired long-term tracking network.