MCUs & MPUs By processing multiple images collected
from externally mounted cameras, the processor is also able to construct a graphic for parking assistance that provides a birds-eye view of the vehicle and surrounding objects (Figure 4). As visual interpretation of scenery is a relatively simple task for the human brain to understand, this can greatly enhance accurate positioning and greatly simplify even complicated parking manoeuvres. Toshiba has developed two Visconti2
processors, the TMPV7504XBG Duetto and TMPV7506XBG Quartetto, offering inputs for two or four high-resolution (up to 1.3 million pixels) colour cameras respectively. Both devices support LDW and FCW functionality, lane-change assistance, traffic sign recognition and adaptive cruise control. The Quartetto can support additional pedestrian detection including backover prevention, and birds-eye view parking assistance. Project development is supported via a Software Development Kit (SDK) with drivers and sample application programs, as well as a dedicated Integrated Development Environment (IDE), media processor debugger and simulator. A reference board populated with a Visconti2 evaluation chip is also available, which helps simplify development and shorten project completion time.
The future
As a powerful image recognition processor, featured for advanced camera-based driver-assistance features, the Visconti2 platform enables developers to provide drivers with the most sophisticated capabilities that can be supported using
Highly reliable, human detection (coHOG Technology)
Figure 4: Processing multiple images to construct birds-eye view for parking
current technologies. As this platform evolves by adding increased processing capability, next-generation ADAS will be able to run additional applications – for example pedestrian detection during daylight – concurrently. Further into the future the integration of network management and safety management features could enable the ADAS to communicate with roadside units implementing detection and wireless technologies. Through systems such as these, tomorrow’s cars may even be able “see” around corners to provide an early warning of a hazard such as a pedestrian crossing the road out of sight of the approaching driver.
Toshiba Electronics Europe |
www.toshiba-components.com
Klaus Neuenhueskes is Senior Marketing Engineer for Automotive System LSI IC Product Marketing at Toshiba Electronics Europe (TEE)
The ability to detect humans within the field of view of a camera-based ADAS is critical to achieving a reliable implementation of features such as pedestrian detection and backover protection, which warns drivers of the presence of bystanders when reversing or, if necessary, initiates vehicle braking. Recognition of the human form is needed in many areas outside of the automotive space, such as security
surveillance, and has been the subject of significant research effort by commercial,
academic and government organisations worldwide. It is a complex field, due to the sheer diversity of the human form, ranging from small children to large adults, as well as the need to detect groups of people. Various types of feature descriptors have been proposed, to aid recognition of human beings. A recent trend is to use descriptors based on the assumption that edges associated with features such as shoulders, arms, legs or hips will have gradients within certain upper and lower limits and they can be detected within specific regions of the anatomy; for
example shoulders and arms will have a certain orientation relative to each other and to the legs, within physiological limits such as maximum and minimum limb lengths, range of movement and overall height. The system is able to detect the presence of a human form by building Histograms of Oriented Gradients (HOG); see illustration.
Shapelets, or combinations of edges, can present a more accurate basis for human detection, but this approach requires more complex processing. To develop reliable
methods of human detection for ADAS applications, Toshiba has been researching a feature descriptor based on multiple-gradient orientation, called Co- occurrence Histograms of Oriented Gradients (CoHOG). In this approach, the building blocks for the histograms are pairs of oriented gradients, which are more descriptive than the single orientation used in a HOG but are less complex than shapelets. Using Toshiba’s CoHOG technique, it becomes possible to implement human detection within a cost-sensitive, low-power embedded application such as an automotive ADAS.
www.cieonline.co.uk
Components in Electronics
April 2012 27
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64