Smart cameras use on-board DSPs for image processing

Dec. 1, 1998
Digital signal-processing (DSP) enabled video cameras can process real-time video data as captured and make decisions based on the contents of the video image. When equipped with a network interface, the cameras can also act as sources of information, rather than merely providing image data. Indeed, such networked digital cameras might spur a variety of new applications in areas from machine vision to security monitoring.

Smart cameras use on-board DSPs for image processing

-Andrew Wilson

Digital signal-processing (DSP) enabled video cameras can process real-time video data as captured and make decisions based on the contents of the video image. When equipped with a network interface, the cameras can also act as sources of information, rather than merely providing image data. Indeed, such networked digital cameras might spur a variety of new applications in areas from machine vision to security monitoring.

Although such "smart" cameras are only available at present from a few vendors, such as Canpolar Technologies (St. John`s, Newfoundland, Canada), DVT (Norcross, GA), and Wintress Engineering (San Diego, CA), other companies are expected to enter the field in a few months. For camera developers, the question of which processor to embed should rely greatly on the amount of support from CPU and DSP vendors.

To ensure its DSP family gains in the smart-camera arena, Texas Instruments (TI; Dallas, TX) demonstrated a C6701-based network camera running a video/graphics setup at the recent Embedded Systems West Conference in San Jose, CA. Developed by Thomas Olson and Leonardo Estevez at TI`s research and development center, the camera comprises a TMS320C6201 processor providing 1600 MIPS of compute power, a video frame grabber, an Ethernet interface, memory, and other peripherals.

System software, based on the SPOX real-time operating system, allows multiple network-camera applications to use the video stream and network interface without conflict and provide a variety of services tailored to support network camera applications. "Currently still a research and development effort, TI does not now offer the network-camera software or the video and network daughterboards as products," says Olson. "While plans for producing the technology are still being finalized, additional announcements are expected," he adds.

In designing the camera, TI chose an approach based on its PCI-based C6201 evaluation module (EVM). Featuring the C6201 processor, 8 Mbytes of DRAM, and the TI EVM peripheral expansion bus, the camera hosts both an EVM-based sensor interface and a network board (see figure). With an on-board NTSC/S video decoder, the sensor interface can capture 640 ¥ 480 ¥ 16-bit images in YUV 4:2:2 format and display them using a VGA RAMDAC. For networking, the stackable network input/output board contains an Ethernet controller with a 10BaseT interface and 2 Mbits of EPROM to support the C6201 boot loader.

In addition, TI researchers are developing a network camera operating system that consists of the SPOX RTOS, a system resource manager, and an application programming interface (API) for video acquisition and display. Moreover, the TI video-analysis API offers developers potential access to high-level functions such as motion detection, object recognition, and tracking. According to Olson, a possible evaluation system could provide developers with a low-cost, easy-to-program evaluation system for machine-vision camera development. By offering such a system to camera developers, TI`s C6000 family would be a primary choice for future "smart" camera designs.

Voice Your Opinion

To join the conversation, and become an exclusive member of Vision Systems Design, create an account today!