To assist drivers and prevent collisions, we propose a system called OmniView that extends driver's vision in all directions, using cameras of multiple collaborating smartphones in the surrounding vehicles. OmniView provides a driver with a traffic map about the relative positions of surrounding vehicles. Under OmniView, each vehicle detects other vehicles in its view, estimates their relative positions, and broadcasts its local map. Upon reception of a map from another vehicle, a vehicle updates its own map by fusing it with the received map. A key issue faced by OmniView is, how does a vehicle address another vehicle in its map? We propose that a vehicle's image itself could be treated as its address. However, if we include images in each map message, the communication overhead will be high. Towards that end, OmniView resolves a vehicle's image to a small unique ID. With this approach, we demonstrate that it is feasible to develop the OmniView system that produces a traffic map in real-time. Besides, through computer vision techniques and the collaboration between vehicles, OmniView could show the positions of surrounding vehicles with reasonable accuracy on the map. Such a traffic map, even without being displayed to the drivers, can act as the common substrate based on which various alerts can be triggered to avoid accidents.