基于表面曲率特征的自适应网格布料建模研究

Research on the Enhancement Algorithm of Qing Dynasty OfficialRobes Detection Based on YoloV5

服装布料模拟可用于服装设计、虚拟现实、游戏影音等领域,具有广泛的应用场景。为了提高布料模拟的效率与真实性,采用自适应网格策略来进行布料模拟,并针对当前自适应网格细分准则只能采用较小的阈值和容易引发过度细分的问题,提出了基于表面曲率特征的自适应细分准则。首先,根据表面曲率的大小对布料网格进行自适应细分;其次,根据曲率变化率对细分层次进行限定,避免产生冗余的精细网格。实验表明:相比于传统的基于曲率的细分准则,采用表面曲率和曲率变化率的双重细分标准可减少冗余网格10%左右,提升仿真速度约15%左右;与基于规则网格的布料模拟相比,仿真时间提升了19%,同时可以获得更好的细分效果。

Real-time detection of the Qing dynasty official robes patch and decorative objects with complex image features will provide an effective reference for the reasonable collocation of relevant film and television characters in time.On this basis,this study constructed a more lightweight network model by introducing the involution operator and other methods in reference to the yoloV5 target detection framework.It can not only share parameters between channels to a certain extent,thereby effectively reduce the computing cost,and meanwhile establish long distance relationship between pixels.Experimental results show that the proposed method is obviously superior to yoloV5 in indicators like mAP,FLOPs,and FPS,both for the target detection of official robes in Qing Dynasty and other detection applications.