PORTLAND, Ore. A smart sensor that aims to lower the cost of automotive, industrial and consumer electronics machine vision systems integrates a DSP chip with a CMOS imager, according to the Swiss Center for Electronics (CSEM), which will unveil its system-on-chip during the International Solid-State Circuits Conference.
The Icycam sensor, the result of nearly a decade of research at CSEM (Neuchatel, Switzerland), is said to build intelligence into firmware, enabling low-cost automotive vision and smart security systems, as well as for optical character recognition.
“Icycam is a new technology based on the combination of our previous CMOS image sensor plus a DSP, and the whole system [is] on one piece of silicon,” said Edo Franzi, CSEM's section head for sensory information processing. “The new image array is QVGA [320 x 240 pixels] and features a digital logarithmic compressor.”
|The “Icycam” combines a CMOS imager (left) with mixed-signal processing that extracts illumination-independent information about contrast and orientation (middle) and a DSP for adding application-specific smarts (right).|
Last year, CSEM demonstrated a two-chip solution adopted for optical character-recognition systems at banks for scanning checks. It is also being used for smart security algorithms that can spot unauthorized activity and issue alerts.
Automobile manufacturers have been designing the sensor into high-end models to detect lane departures, pedestrians, cars in drivers' blind spot and curbs for automatic parking functions. However, the single-chip device could lower the price of smart vision systems enough to make them standard equipment on future automobiles, the developers claim.
According to ABI Research, as many as 3 million automobiles could include smart vision chips by 2012 if inexpensive smart-vision chips perform as advertised.
A new smart compression algorithm in the CSEM's SoC uses mixed-signal processing to simplify the extraction of features from real-time video streams. The digital logarithmic compressor algorithm extracts the edges and outlines of objects in a scene that are independent of illumination levels. This is achieved by changing the pace of integration as images are acquired.
The system works by adding up the photocurrent at each pixel on separate capacitors at a rate that changes on a logarithmic scale. Compression is automatically achieved since each pixel logarithmically changes the amount of charge transfered to its capacitor for each incident photon over the span of the integration time.
The final charge value for each pixel is then run through a state machine (used to model the behavior composed of a finite number of states), that outlines objects in a scene. The calculations are enabled by the logarithmic encoding at the pixel level, which enables simple arithmetic operations among neighboring pixels to calculate contrast and the direction of image features (orientation).
The resulting image, which is rendered identically regardless the illumination level, is passed to the on-chip DSP, which performs application-specific image processing steps such as sensing other cars for automotive applications, motion for security applications or handwritten characters on checks.