A Study on Pedestrian Re-identification Based on Transfer Learning

In recent years, with the continuous development of deep learning-based image classification and recognition, artificial intelligence has become the most representative keyword of our time, intelligent video surveillance technology has also become a current research hotspot. Intelligent video surveillance system is mainly developed on computer vision, pattern recognition and machine learning techniques. It first captures videos or images by the surveillance camera, then filters redundant data, and finally extract effective features from the data for accurate detection and analysis. Pedestrian detection and Person re-identification (ReID) are the core of intelligent video surveillance systems. However, many negative factors such as the lack of large-scale datasets, the minor differences between pedestrians and the complex surveillance environment, still pose considerable challenges to the pedestrian re-identification task. In this paper, we proposed a deep convolutional neural network for video pedestrian re-identification algorithms. The proposed framework is developed based on the motion trajectory of pedestrian. Concretely, we first design a good deep convolutional neural network and train it using transfer learning technique, then we employ it and a weighted probability-based decision mechanism to recognize pedestrians in a series of video frames. To verify the effectiveness of the proposed method, we conduct relative experiments on the public dataset. The experimental results show that the proposed method achieves good performance in pedestrian re-identification.