Title: TRACKING PEOPLE AND CARS USING 3D MODELLING AND CCTV

Tracking people and cars using 3D modelling and CCTV

Gerda Edelman, NFI 

Introduction

In forensic casework, CCTV footage can provide useful information about the crime, perpetrator or witnesses. However, with the growing number of security cameras, the amount of information increases rapidly. The question arises if surveillance images could be used more effectively with the help of 3-dimensional models of the scenes that are visible in the surveillance images. 

At the Netherlands Forensic Institute a project is being carried out that aims at the 3D reconstruction of all movements of people and cars before, during and after a big incident from analysis of all available video footage. Methods developed in this project are being applied in forensic casework. A case example is presented in this chapter.  

Figure 5: Images of six different security cameras, confiscated by the police 

Case example

After a shooting incident, police investigators confiscated footage of six private security cameras in the surrounding area, from each of which a single frame image is shown Figure 5. On these videos many people and cars were visible, all of whom were of potential interest. For scenario testing and the validation of testimonies a 3D reconstruction of all movements was made.  

First, a 3D model of the scene was made by aerial stereo photography, in which virtual cameras were created at the positions of the real security cameras (see figure 6 for a top view). The virtual camera parameters (location, focal length, orientation) were adjusted until the virtual camera view was close to the real camera view. An example of a camera view and its virtual imitation is shown in Figure 7. There was some overlap between the different camera views.  

Figure 6: Top view of the 3D model showing the area around the crime scene. Cones demonstrate fields of view of the security cameras. 

Figure 7: Example of a camera image and its virtual imitation. At the position of a car a box is created 

In the 3D model, boxes and cylinders were created at the positions of cars and persons that were visible on the camera images (see Figure 7). When the car/person moved, the box/cylinder was moved to the same location. In this way a 3D animation was created of all movements observed in the area of surveillance, which could be watched from different perspectives.  

Discussion

The 3D model allows more insight into the situation because an overview of the scene can be given. In this case, the 3D reconstruction gave insight in the relations between cars seen on different cameras. Because the fields of view of the cameras partly overlapped on the streets, cars could be tracked over six different cameras. However, when a car went out of sight and came back later, it was not possible to identify it again, because of the low quality of the images. On the path areas there was hardly any overlap between different camera views. This, in combination with the low resolution of the images and the size of a person, made it more difficult to track pedestrians on multiple cameras. Nevertheless, more awareness could be created of the locations of people because the scene could be watched from any point of view. This made it possible to verify different scenarios and testimonies.  

Another advantage of this method is that it can be combined easily with other spatio-temporal information. Cell sites, for example, can be added to the 3D model, so that evidence concerning telephone calls can be compared with the positions of different people at the same time.  

Also, the analysis is objective and can be compared to results of other investigators. Differences in interpretations can be visualized and separated from facts in the animation. This is of high importance in forensic investigations.