How much data does an automotive camera really need to collect? Must it provide high-resolution pictures to the system using its input, or is a less data-intensive output enough? These questions drove the folks at CSEM (Neuchâtel, Switzerland; www.csem.ch) to develop a nonconventional 160 x 128-pixel vision sensor that reduces the amount of data delivered, and should reduce the cost of in-vehicle vision systems.
According to Dr. Christian Enz, v.p. Microelectronics at CSEM, “Each feature—contrast, direction, and luminance—can be output independently for each frame, and the chip does an on-board analog computation of magnitude and direction of image contrasts.” As a result, the sensor immediately adapts to rapidly changing light conditions, and outputs a somewhat grainy negative of the scene its lens is pointed at. “It is easy to determine from this image what is important and what is not because the contrast between objects is quite obvious. The extraneous information in the image is removed which not only reduces the computing necessary on the chip, but that required by the systems using this information to make decisions,” claims Enz. This sensor is part of the Devise-E low-cost vision system that CSEM hopes suppliers and OEMs will consider for next-generation safety systems.—CAS