Implementation of Multi-object Recognition Algorithm Using Enhanced R-CNN


  • Hyochang Ahn
  • June-Hwan Lee
  • Han-Jin Cho


Background/Objectives: Multi-object recognition is emerging as a technology that can be applied to various real worlds such as image security, gesture recognition, robot vision, and human robot interaction, and it is difficult to recognize public objects in a complex background. Methods/Statistical analysis: Most multi-object tracking methods suffer from performance degradation due to the number of objects changing each frame, and this phenomenon is more pronounced in complex backgrounds. Therefore, in this paper, we propose an improved R-CNN-based multi-object recognition method that can detect multiple objects quickly while being robust to geometric distortions, lighting changes, and noise of environmental elements and objects in the image. Findings: Experiments were compared the detection rate and detection rate of the CNN method, R-CNN method and the proposed method. The proposed method takes more time to recognize multiple objects, but the object recognition rate shows a high result. Improvements/Applications: Through the proposed deep learning based multi-object recognition, it can contribute to the research that monitors and tracks several objects at the same time in the surveillance system.