Mar 2 2015
Toshiba Corporation today announced the development of TMPV7608XBG, an image recognition processor that provides high accuracy recognition and detection of vehicles and pedestrians at night. This sample shipment started in January. This performance is achieved with a device that can carry out 1.9 Tera operations per second (TOPS) and that integrates four new feature descriptors that make use of color-based image information.
The development was announced at the 2015 IEEE International Solid-State Circuits Conference (ISSCC 2015) in San Francisco, California, on February 25.
High-performance processing performance, 1.9 TOPS at peak performance with 3.37 Watts power consumption by the image recognition engines (MPEs) and the accelerators, is due to a heterogeneous architecture that integrates eight media processing engines (MPEs) for floating-point arithmetic processing and 14 image processing accelerators, including the new ones mentioned below.
The image recognition accelerators in the TMPV7608XBG supports four sets of feature descriptors based on color-based image information; Color-CoHOG[1], CoHED[2], CoHD[3] and Color Histogram, in addition to CoHOG[1], Toshiba original image recognition descriptors used by earlier products. Integration of these sets of feature descriptors in TMPV7608XBG boosts recognition accuracy, especially at night and locations with low differences in luminance between objects and the background. TMPV7608XBG integrates two new accelerators that make use of color-based information for image recognition and that offer high-speed image recognition at a rate about 4 times faster than earlier accelerators, effectively shortening the recognition time for various objects.
As a result, TMPV7608XBG offers a single-chip solution able to simultaneously perform the following applications: Traffic Light Recognition (TLR), Traffic Sign Recognition (TSR), Lane Departure Warning (LDW) and Lane Keeping Assist (LKA), Vehicle and Pedestrian Collision Warning and Collision Avoidance, High-Beam Assistance, and General Obstacle Collision Warning.
The importance of Advanced Driver Assistance System (ADAS) is increasing as a solution offering safer driving with less stress, and driverless cars are an object of intense research. Moreover, the new European New Car Assessment Programme, Euro NCAP, will come into force in 2018 and pedestrian collision avoidance during daytime and at night is also scheduled for evaluation of image recognition processors with the high recognition performance are required to meet requirements and targets. It is expected that applications required for the future Advanced Driver Assistance System will continue to increase and require improved processing performance. Toshiba is responding with focused research and development and by increasing the lineup of the TMPV760 series.
[1] CoHOG: Co-occurrence histograms of oriented gradients
[2] CoHED: Co-occurrence histograms of pairs of edge orientations and color differences
[3] CoHD: Co-occurrence histograms of color differences