Sensing Shared Spaces using 3D Stereo-Vision sensors
This MSc project is conducted by Rishabh Mittal, under the supervision of Yufei Yuan, Winnie Daamen and Serge Hoogendoorn from TU Delft, Thomas Paul and Laurens Tait from Arup B.V.. The main idea is to test the data-gathering capabilities of 3D stereo-vision sensors in observing pedestrians and cyclists in a shared space environment.
In this research, we are combining 3D stereo-vision technology with a machine learning (ML) based object detection algorithm. The 3D camera records depth information in addition to visual information from the scene (see an example in Figure 1). Next, ML-based object detection is used to identify people based on their mode of travel (pedestrian/cyclists). Combining both technologies enables us to extract 3D coordinates of each person with every time-step, thus resulting in trajectories.
To test this new data collection setup, two experiments have been planned. The first experiment was performed in a controlled environment within the TU Delft campus – Green Village. We opt to challenge the setup by re-creating certain aspects of shared spaces and crowded situations. As seen in Figure 1, the left image contains both pedestrians and cyclists sharing the same space, and the right image contains a depth map computed by the 3D camera.
The second experiment aims to test the setup in real-world conditions. On 17th October, the field experiment took place in the shared space behind Amsterdam Central Station, near the ferry waiting area. The 3D sensor was mounted on a MCSAVI (multi-camera stand-alone video installation) pole of 6 m high, monitoring traffic flow movement in the shared space area, as shown in Figure 2. The collected video footage needs to be further processed and analyzed to obtain all information of motion of individual pedestrians and cyclists (trajectories, directions, dwell time, speed, density and flow), which in turn can be used to validate the data collection methodology, and to facilitate related research using this sensor, e.g., to reveal traffic flow behaviour in shared spaces.
