Multi-RPN Fusion-based Sparse PCA-CNN Approach to Object Detection and Recognition for Robot-aided Visual System

Object detection and recognition plays an important role in robot-aided visual system. However, when the object's scale varied largely, current object detection algorithms based on convolutional neural network (CNN or ConvNet) (i.e. Fast R-CNN or Faster R-CNN) always cause a high misdetection rate and a low computation speed. In addition, these methods also have the inherent problem: over-fitting. Therefore, we proposed a new multi-RPN fusion-based sparse PCA-CNN detection algorithm to solve the above problems. Firstly, a multi-RPN (Region proposal network) fusion method is adopted to generate candidate windows. Then the new sparse PCA-CNN algorithm is proposed to detect and recognize objects from images or video sequence. Finally, an experiment is adopted to verify the proposed algorithm. And, compared with the existing methods, the results indicate that the proposed method has a better performance.