Nowadays, autonomous driving car is not a dream any more. To make them accessible to general users take a lot of efforts in terms of increasing security and reducing cost. This work I conducted at MERL during a summer intern focused on how to use a single camera to help the car detect obstacles in the front or in the back.
The videos shows the configuration of our system which essentially just a single camera and the output of our system, a sequence of dense 3D reconstruction of the scene. The challenge of the project is that conventionally we compute depth out of stereo image pair, but we enable depth computation using only a single monocular camera.