BoxCars: Improving Fine-Grained Recognition of Vehicles using 3-D Bounding Boxes in Traffic Surveillance

• https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8307405
• Publication web page HTML

fine-grained recognition, traffic surveillance, 3D bounding boxes, convolutional neural networks

In this paper, we focus on fine-grained recognition of vehicles mainly in traffic surveillance applications. We propose an approach that is orthogonal to recent advancements in fine-grained recognition (automatic part discovery, bilinear pooling). Also, in contrast to other methods focused on fine-grained recognition of vehicles, we do not limit ourselves to a frontal/rear viewpoint, but allow the vehicles to be seen from any viewpoint. Our approach is based on 3D bounding boxes built around the vehicles. The bounding box can be automatically constructed from traffic surveillance data. For scenarios where it is not possible to use precise construction, we propose a method for an estimation of the 3D bounding box. The 3D bounding box is used to normalize the image viewpoint by "unpacking" the image into a plane. We also propose to randomly alter the color of the image and add a rectangle with random noise to a random position in the image during the training of Convolutional Neural Networks. We have collected a large fine-grained vehicle dataset BoxCars116k, with 116k images of vehicles from various viewpoints taken by numerous surveillance cameras. We performed a number of experiments which show that our proposed method significantly improves CNN classification accuracy (the accuracy is increased by up to 12 percentage points and the error is reduced by up to 50% compared to CNNs without the proposed modifications). We also show that our method outperforms state-of-the-art methods for fine-grained recognition.