Implementation considerations for single-camera steering assistance systems on a fixed point DSP

Abstract

The design flow of camera-based steering assistance algorithms usually begins with their implementation in floating-point on a PC or workstation. This abstraction from all implementation effects allows an exploration of the algorithm space. Memory, throughput and word-length requirements may not be important issues for offline implementation of the algorithms, but they can become critical issues for real-time implementations on embedded processors. The implementation of driver assistance systems is faced with practical constraints because these algorithms usually need to run in real-time on fixed point Digital Signal Processors (DSPs) to reduce total hardware cost.In this paper we first evaluate numerical requirements for implementation of camera-based lateral position detection algorithms, such as Lane Keep Assistant and Lane Departure Warning.We then present methods that address the challenges and requirements of fixed-point design process. The flow proposed is targeted at converting C/C++ code with floating-point operations into C code with integer operations that can then be fed through the native C compiler for a fixedpoint DSP. We demonstrate the flow on tracking example (Extended Kalman Filter) using synthetically generated data, and we analyze trade-offs for algorithm implementation in fixed-point arithmetic.

Exploration of FPGA-based dense block matching for motion estimation and stereo vision on a single chip

“…However, these devices come with a high energy consumption and their complex memory and cache models prohibit the calculation of bounded execution times and latencies, which is crucial for safety-relevant applications. DSPs represent a third alternative and have been successfully utilized in automotive vision in the past [6], [7], but also suffer from their fixed architectural template, which is not adaptable to higher application complexity. In contrast to these hardwired processors, FPGAs provide the opportunity to adjust the device to the application structure, which enables an efficient and optimized acceleration of computation-intensive parts [8].…”

“…For a camera-based system, all latencies are not considered critical, as the image capturing process for a complete frame usually requires 16-40 ms, which dominates the overall application latency. However, all variants without run-time reuse (# 1,2, 6,9) show latency advantages, which is explained by the early availability of the final processing results, which must not be time-multiplexed with later matches.…”

Algorithmic and software techniques for embedded vision on programmable processors

“…In this case techniques for algorithm conversion from floating-point to fixed-point arithmetic mentioned in Section 4.1 can help significantly reduce the execution time. Good candidates for such considerations are Jacobian matrix calculations and observation updates, as described by Nikolić [28].…”