Abstract—This paper describes a system for precise vehicle retrieval in traffic surveillance, exploiting image representation based on Convolutional Neural Network. Our training data contains about 450k vehicle face images cropped from different scene and angle. We prepare two test datasets, one set of images all from daytime and another set of images in night scene. Experiment results on data from a real city traffic surveillance network validate and evaluate our system. We demonstrate that our CNN features have outstanding performance in both two test-sets and achieve an expected recall of 97.89% and precision of 58.22% at top100 for the first test-set and recall of 97.09% and precision 57.57% for the second one. Our work is very import especially in traffic surveillance and public surveillance to find one special vehicle.
Index Terms—Precise vehicle retrieval, convolutional neural network, siamese structure.
The authors are with the Internet of Things Technology Department, the Third Research Institute of the Ministry of Public Security, Shanghai, P.R. China (e-mail: uestcsby@163.com).
[PDF]
Cite:Boyang Su, Jie Shao, Jianying Zhou, Xiaoteng Zhang, Lin Mei, and Chuanping Hu, "The Precise Vehicle Retrieval in Traffic Surveillance with Deep Convolutional Neural Networks," International Journal of Information and Electronics Engineering vol. 6, no. 3, pp. 192-197, 2016.
