Automatic Camera Calibration for Traffic Understanding

Abstract

Figure 1: We automatically determine 3 orthogonal vanishing points, construct vehicle bounding boxes (left), and automatically determine the camera scale by knowing the statistics of vehicle dimensions. This allows us to measure dimensions and speed (right) and analyze the traffic scene. This paper proposes a method for fully automatic calibration of traffic surveillance cameras. Our method allows for calibration of the camera -including scale -without any user input, only from several minutes of input surveillance video. The targeted applications include speed measurement, measurement of vehicle dimensions, vehicle classification, etc.The first step of our approach is camera calibration by determining three vanishing points defining the stream of vehicles (Fig. 2, [3]). The second step is construction of 3D bounding boxes of vehicles (Fig. 3) and their measurement up to scale. In the third step, we use the dimensions of the 3D bounding boxes for calibration of the scene scale ( The next step of our approach is construction of 3D bounding boxes of the observed vehicles (Fig. 3). We assume that the vehicle silhouettes can be extracted by background modeling and foreground detection and that the vehicles of interest are moving from/towards the first vanishing point. The 3D bounding box is constructed using tangent lines from vanishing points to the blob's boundary.Having the bounding box projection, it is directly possible to calculate the 3D bounding box dimensions (and position in the scene) up to precise scale. By fitting the statistics of known dimensions and the measured data from the traffic, we obtain the scale of the scene (Fig. 4).Camera orientation together with a know distance enables for measuring of vehicle speed/size or distances in the scene. We measured several

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

“…However, we propose to use several data normalization and augmentation techniques to significantly boost the classification performance (up to 50 % error reduction compared to base net). We utilize information about 3D bounding boxes obtained from traffic surveillance camera [49]. Finally, in order to increase the applicability of our method to scenarios where the 3D bounding box is not known, we propose an algorithm for bounding box estimation both at training and test time.…”

“…The method works with arbitrary viewpoints and we require only 3D bounding boxes of vehicles. The 3D bounding boxes can either be automatically constructed from traffic video surveillance data [49], [50] or we propose a method on how to estimate the 3D bounding boxes both at training and test time from single images (see Section III-D).…”

“…In traffic surveillance applications, it is common that prior fine-grained vehicle classification is necessary to detect vehicles; therefore, we include a brief overview of existing methods for vehicle detection. It is possible to use standard [49]. The image was adopted from the paper with the authors' permission.…”

“…We based our work on 3D bounding boxes proposed by [49] ( Fig. 4) which can be automatically obtained for each vehicle seen by a surveillance camera (see Figure 2 for schematic 3D bounding box construction process or the original paper [49] for further details). These boxes allow us to identify the side, roof, and front (or rear) side of vehicles in addition to other information about the vehicles.…”

“…As proposed by [49], the vehicle's contour and vanishing points are required for the bounding box construction. Therefore, it is necessary to estimate the contour and vanishing points for the vehicle.…”