Footnotes

While most of the scene is immutable and can be understood based on GPS (provided satellite signal), neighbouring cars are unpredictable and cannot be studied off-line. Human behaviour is hard to simulate.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... anomalies ²

Mobile GPUs are still in the stage of gradual acceptance in the market, but they are consistently expensive.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... experiment ³

The equivalent of synthetic data, or the applications run on known data with simple characteristics.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...combined ⁴

The speed of digestion of frames is at around 12 per second (maximum) for the hardware and conditions at hand (depending on light, available RAM, etc.), but when some more processing comes into play it slows down the actual display rate (not capturing rate) down to about 0.1, depending on what the program is trying to achieve. I have put the tablet in debugging mode so that I can view system statistics (load average and other kernel-level output) on the tablet's screen, along with real-time shell output (warnings for example) in Eclipse running Ant.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... available ⁵

The script can be handy later on when assessing performance elsewhere in the algorithm.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...⁶

VGA being processed on single-core Rockchip 2918 CPU running at 1 GHz with 1024 MB of RAM, Linux kernel version 3.0.8, and Android ICS 4.0.3 to be more specific

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.