Vehicle counting using computer vision techniques has potential to alleviate traffic congestion in intelligent transportation system. In this paper, we propose a novel method to count vehicles in a human-like manner. This paper has two main contributions. Firstly, we propose ST-CSNN, which is an efficient, lightweight vehicle counting method. The method counts based on vehicle identity comparison to omit duplicate instances. Combined with the spatio-temporal information between frames, it is able to accelerate speed and improve accuracy of counting. Secondly, we strengthen the method’s performance by proposing an improved loss function on the basis of Siamese neural network. Besides, we conduct experiments on several datasets to evaluate the performance of the proposed loss function for verification and the whole method for counting. The experimental results show the practicability of this method for real counting scenes.

使用计算机视觉技术进行车辆计数具有缓解智能交通系统中交通拥堵的潜力。在本文中，我们提出了一种以类人方式计算车辆的新方法。本文有两个主要贡献。首先，我们提出了 ST-CSNN，这是一种高效、轻量级的车辆计数方法。该方法基于车辆身份比较进行计数以省略重复实例。结合帧间的时空信息，可以加快计数速度，提高计数精度。其次，我们通过在 Siamese 神经网络的基础上提出改进的损失函数来增强方法的性能。此外，我们在几个数据集上进行了实验，以评估所提出的用于验证的损失函数和整个计数方法的性能。