E-con’s “TaraXL” stereo vision camera works with Nvidia’s Linux-driven Jetson TX2 module and other Tegra based systems to stream 3D depth maps with 752 x 480 resolution at up to 50fps over USB 3.0.

E-con Systems’ TaraXL is designed for stereo vision imaging applications including autonomous driving, robotics, drones, mixed reality applications, people detection/counting, stock level monitoring, volume measurements, and proximity warning systems. Like the very similar, two-year old Tara camera, the TaraXL has a USB 3.0 interface and is built around a pair of OnSemi 1/3-inch MT9V024 CMOS image sensors, which enable WVGA Global shutter monochrome image capture. The main difference is that the TaraXL SDK is five times faster than the Tara SDK, claims E-con.

TaraXL alone (left) and with Jetson TX2 Developer Kit
(click images to enlarge)

Whereas the original Tara camera’s SDK is built on OpenCV, and available for Windows and Linux desktops, the TaraXL SDK leverages Nvidia’s CUDA libraries in addition to OpenCV. The camera requires a computer with Nvidia Tegra processors, and is optimized for the Linux-driven Nvidia Jetson TX2 module and its Tegra SoC’s 256-core Pascal GPU.

The camera’s micro-USB 3.0 connection is intended primarily to link up to the Jetson TX2 Developer Kit, but can also work with other Tegra based systems. The camera enables streaming of uncompressed Stereo WVGA (1504 x 480), delivered as dual 752 x 480 images at up to 60fps. The TaraXL SDK provides 3D depth maps at 752 x 480 resolution at up to 50fps “without stressing the CPU,” says E-con.

The TaraXL SDK also supports a high accuracy 25fps depth mapping mode. Other options include cropped VGA and QVGA (640 x 420), both at 60fps.

The MT9V024 sensors offer 8- or 10-bit monochrome capture and an enhanced NIR pixel process that provides “excellent image quality” in IR illuminated applications, says E-con. The sensors have a 50cm to 300cm depth range and are separated by an inter-ocular base line distance of 60mm. The fixed-focus, pre-calibrated lens pair is mounted on S-Mount (M12 lens mount) holder.

The camera ships with a micro-USB to USB host cable and provides backward compatibility with USB 2.0 host ports. The TaraXL does not require any special camera drivers to be installed, says E-con. The 100 x 35 x 30mm camera supports -30 to 70°C temperatures and weighs 80.5 grams with the enclosure and 28.5 grams without.

TaraXL front and back views (left) and TaraXL Studio sample application
(click images to enlarge)

The TaraXL SDK is built on top of OpenCV 3.1.0 and Nvidia CUDA Compute Capability 6.2. The SDK supports disparity and depth maps, which can be used to develop autonomous roving and machine vision applications. TaraXL SDK natively comes with C++ bindings that can be used to build wrappers for platforms such as ROS (Robot Operating System).

NileCAM30

Linux sample applications with source code are available in the bundled TaraXL Studio program. The apps demonstrate synchronous stereo image streams, disparity maps, and depth measurements. Customization services include baseline customization and sensor or optics/lens change services. Software assistance is also available.

Demo of an autonomous warehouse rover bot running on Jetson TX2 with TaraXL Stereo Camera
Source: E-con Systems

Other E-con Systems embedded cameras designed to work with the Jetson TX2 module include the 3.4-megapixel NileCAM30 GMSL camera, which provides both USB 3.1 and MIPI-CSI camera connections. Other E-con MIPI-CSI based camera kits that support both the TX1 and TX2 include the e-CAM120_TRICUTX2, which features three 13-megapixel, [email protected] cameras and the e-CAM30_HEXCUTX2 with 6x 3.4-megapixel HD cameras.

SLAM mapping with TaraXL

 
Further information

The TaraXL stereo vision USB 3.0 camera is currently available for evaluation at a special launch price of $299 (normally $349) through Sep. 13. Volume discounts are available. More information may be found on E-con Systems’ TaraXL announcement, product page, and shopping page.