采用模块化技术，设计一种机器人智能分拣系统。在该平台下，拟提出一种新的变化检测算法-LBPSG算法，利用改进的基于局部二值模式的变化检测算法与分离高斯模型变化检测算法进行融合以提高实时性；拟采用一种改进型粒子滤波算法为LBPSG算法提供辅助信息，用于实现对兴趣物体变化检测，以及实现当兴趣物体出现部分遮挡、重叠时的有效检测；在三维点云目标识别与理解部分，重点改进基于局部特征描述子算法RoPS，实现对多目标场景的实时识别与理解，构建可供机器人作业推理的三维语义地图描述文件；改进CBR-BDI推理机制，通过语音交互、反馈引导，得到用户期望，将用户期望结合实时场景进行分析、推理，生成用户的最终意图，形成具有语义信息的解决方案，最终实现模块化机械臂的智能分拣作业。针对系统的整体实时性，对体系结构及改进算法的性能进行实际的验证、测试、分析与评价,完成一款具有高度展示度的模块化机械臂智能分拣系统平台。

A kind of 3D scene reasoning and interaction algorithm for sorting system of modular manipulator is proposed in the project, and the hardware and software platform will be designed. At first, a new change detection algorithm (LBPSG) is proposed to improve the system real-time performance, which will use the LBP (Local Binary Pattern) with Split Gaussian Models, and an improved Particle Filter Algorithm will be used to provide supplementary information for LBPSG algorithm to implement the detection of ROI (Regions Of Interest) object changing, even if target objects are overlapping or partly occlusive. And then the research work will focus on the improvement of local feature descriptor RoPS which is the key point in the proceeding of 3D point cloud object real-time recognition and understanding, to construct 3D semantic map description file for robot to reason. Lastly, an improved CBR-BDI reasoning mechanism will be proposed, user’s desire will be got through voice interaction and feedback guidance, combining with real-time scene, the system will analysis the desire to reason, generate the final intention and the semantic solution, achieve intelligent sorting by using modular manipulator. At all, the real-time performance of the whole system will be resulted from the actual verification, testing, analysis and evaluation of the system architecture and all improved algorithms, and a high quality demonstration platform of intelligent sorting system using modular manipulator will be completed.