Recently, millimeter-wave (mmWave) radar has been widely used in many applications, such as advanced driver-assistance systems (ADAS) on vehicles and short-rangeremote sensing (e.g., road traffic monitoring and elderly health monitoring), due toits advantages including robustness to adverse weather and illumination conditions,privacy compliance, high resolution and compact size, compared to the other sensingcounterparts. In principle, the mmWave radar can measure the range, angle andDoppler (radial velocity) of moving objects in a scene. With a chain of detection,clustering and tracking algorithms, a radar point cloud can be obtained to offer theinformation including location, velocity and trajectory of objects. However, objectclassification or recognition from its radar point cloud is a major problem that needsto be solved to meet the latest demands in those advanced applications. In this dissertation, we leverage the recent developments in deep learning to explore the solutionsof radar point cloud classification from supervised, unsupervised and semi-supervisedapproaches. Particularity, we proposed the Hybrid Variational RNN Autoencoder(HVRAE), as a generative model, to detect the elderly fall from the radar pointcloud, among other applications such as traffic monitoring and patient behavior classification.
On the other hand, as the mmWave radars have increasingly been used widely, theinterference among these radars will become a severe problem in the near future. Tomitigate the radar interference, we proposed an adaptive noise canceller (ANC) basedsolution that can increase the Signal-to-Interference Ratio (SIR) of targets at a lowcost. Both the simulation and experiment show excellent SIR improvement.
猜您喜欢
推荐内容
开源项目推荐 更多
热门活动
热门器件
用户搜过
随便看看
热门下载
热门文章
热门标签
评论