MCUs & MPUs
Embedding safer driving in silicon
Klaus Neuenhueskes explains how high-performance, multi-parallel image processing is allowing camera- based Advanced Driver Assistance Systems (ADAS) to support enhanced safety features
I
mproving road safety has become one of the most important issues influencing new vehicle design. As airbags, crumple zones and passenger protection cells have become standard on almost all vehicles, attention is switching to implementing more effective systems to help drivers avoid accidents altogether. In particular, a great deal of research has been devoted to developing Advanced Driver Assistance Systems (ADAS). Typically beginning in prestige vehicles, various types of ADAS are now permeating manufacturers’ vehicle ranges. Awareness of objects and events in the immediate vicinity of
as measurement of driver ‘distraction’ – e.g. if the driver is not looking at a potential hazard a warning signal can be given or braking can be initiated earlier. Finally, while most or all of these features
Figure 1: Vehicle safety features using camera-based ADAS
the vehicle is critical to the operation of ADAS. A number of detection techniques are viable, such as front-mounted radar for Forward Collision Warning (FCW). However, camera-based detection holds some advantages for FCW, and for other ADAS features. A central forward-looking camera mounted high-up in the cabin can provide detection for Lane Departure Warning (LDW) minimising the number of cameras and camera input channels needed. In addition camera-based systems can present a direct visual representation of the detected scene on a dashboard display, and further image processing can enhance the displayed picture with overlaid graphics.
An example of the use of image enhancement is to draw attention to nearby pedestrians that otherwise may not be clearly visible to the driver. This can be a powerful safety tool for the driver, for example when reversing or when visibility is poor. Image-based sensing also enables a number of driver-related safety features via driver-facing cameras that monitor face orientation. These include drowsiness detection and driver authentication, as well
26 April 2012
could be implemented using monochrome cameras, the use of high-resolution colour cameras provides the opportunity to develop traffic signal recognition. Figure 1 summarises ADAS features that are currently possible using camera-based technology.
Optimised ADAS image processing Advanced camera- based ADAS is highly dependent on effective image recognition. Key demands include high accuracy with 100% hazard detection and zero false recognition, which are necessary in order to maintain driver
relatively quickly. And to meet these needs ADAS developers are demanding high- performance, dedicated embedded processor solutions that can manage all activities: image capture, pre-processing, features extraction, classification, post- processing and information display. Among the latest examples of this type of embedded solution is the Visconti 2 specialised image recognition processor. This device combines multi-parallel media processor architecture and multiple image- processing accelerators with a two- or four-channel camera interface, 32-bit control CPU and communication, display and memory peripherals (see Figure 2). Its architecture takes advantage of the Venezia media processor from Toshiba, which comprises four Media Processing Engines (MPEs) each containing a 32-bit RISC CPU and co-processor plus I- and D- cache and integrated high-speed RAM. Processing data from the MPEs and accelerators the 32-bit control CPU creates an interpretation of the driver environment. Because the device is capable of managing image processing for multiple functions, it can be used to manage a variety of ADAS capabilities. These include lane, vehicle and pedestrian detection and traffic sign recognition. The integrated graphic processor and display unit allows for the output of images with overlaid warnings to the driver display. Integrated alongside the multi-parallel
MPE core, are a number of image processing accelerators for frequently used
The use of wide-angle lenses in some ADAS applications enables the system to monitor a wide field of view with a single camera, but demands post-processing to eliminate distortion from the image displayed for the driver. In the case of Visconti2, an Affine Transform Accelerator is used to perform manipulations such as resizing, rotation and shear to accomplish this. In addition, two filter accelerator
Figure 3: Pedestrian detection and marking
channels perform noise reduction, smoothing, edge detection and colour correction.
In addition to these accelerators, three further accelerators are implemented in some variants to support pedestrian detection capability. Pedestrian detection in the processor is performed using a technique known as the Histogram of Oriented Gradients (HOG) methodology. HOG compares processed detected data against known characteristics of human body shapes and movement to distinguish pedestrians from other objects that may be near the road, such as posting boxes or road signs.
Figure 2: Multi-parallel and hardware-accelerated image processor for ADAS
confidence in the system. High-speed computation of results is also imperative, especially for FCW and pedestrian recognition as the vehicle can be approaching such slow moving hazards
Components in Electronics calculations such as distortion
compensation and image enhancement. In addition, advanced acceleration functions embody state-of-the-art IP for pedestrian detection.
The driver’s view The highly integrated Visconti image recognition processor implements a display interface allowing direct connection to a dashboard LCD display. Images presented to the driver are corrected for distortion, and the processor has the ability to mark on the image to highlight features such as lane boundaries as well as pedestrians detected in the field of view (Figure 3). This marking ability complements the advanced CoHOG human detection implemented in the dedicated CoHOG accelerator block to provide a clear warning of the presence of pedestrians in front or behind the car even in low-contrast situations. In some cases such pedestrian detection may go beyond warnings and actually initiate autonomous braking if an accident becomes unavoidable.
www.cieonline.co.uk
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