Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AppSign: Multi-level Approximate Computing for Real-Time Traffic Sign Recognition in Autonomous Vehicles

Fatemeh Omidian, Athena Abdi

TL;DR

In AppSign a novel approximate multiplication method called TIRuD is proposed that truncates the operations while keeping the accuracy acceptable, and provides the adaptive approximation of the underlying CNN by involving various levels of computation and considering different approximation methods.

Abstract

This paper presents a multi-level approximate computing approach for real-time traffic sign recognition in autonomous vehicles called AppSign. Since autonomous vehicles are real-time systems, they must gather environmental information and process them instantaneously to respond properly. However, due to the limited resources of these systems, executing computation-intensive algorithms such as deep-learning schemes that lead to precise output is impossible and takes a long time. To tackle this, imprecise computation schemes compromise the complexity and real-time operations. In this context, AppSign presents a multi-level approximate computing scheme to balance the accuracy and computation cost of the computation-intensive schemes and make them appropriate for real-time applications. AppSign is applied to the CNN-based traffic sign recognition unit by approximating the convolution operation of CNN which is the primal solution for image processing applications. In AppSign a novel approximate multiplication method called "TIRuD" is proposed that truncates the operations while keeping the accuracy acceptable. Moreover, it provides the adaptive approximation of the underlying CNN by involving various levels of computation and considering different approximation methods. The efficiency of the proposed AppSign, in real-time traffic sign recognition, is evaluated through several experiments. Based on these experiments, our proposed TIRuD reduces the accuracy by about $10\%$ while saving execution time about $64\%$ over the exact multiplication, averagely. Moreover, employing our proposed hierarchical approximation in various model layers outperforms the exact computation $27.78\%$ considering "AoC" that joins accuracy and computation cost in a parameter.

AppSign: Multi-level Approximate Computing for Real-Time Traffic Sign Recognition in Autonomous Vehicles

TL;DR

In AppSign a novel approximate multiplication method called TIRuD is proposed that truncates the operations while keeping the accuracy acceptable, and provides the adaptive approximation of the underlying CNN by involving various levels of computation and considering different approximation methods.

Abstract

This paper presents a multi-level approximate computing approach for real-time traffic sign recognition in autonomous vehicles called AppSign. Since autonomous vehicles are real-time systems, they must gather environmental information and process them instantaneously to respond properly. However, due to the limited resources of these systems, executing computation-intensive algorithms such as deep-learning schemes that lead to precise output is impossible and takes a long time. To tackle this, imprecise computation schemes compromise the complexity and real-time operations. In this context, AppSign presents a multi-level approximate computing scheme to balance the accuracy and computation cost of the computation-intensive schemes and make them appropriate for real-time applications. AppSign is applied to the CNN-based traffic sign recognition unit by approximating the convolution operation of CNN which is the primal solution for image processing applications. In AppSign a novel approximate multiplication method called "TIRuD" is proposed that truncates the operations while keeping the accuracy acceptable. Moreover, it provides the adaptive approximation of the underlying CNN by involving various levels of computation and considering different approximation methods. The efficiency of the proposed AppSign, in real-time traffic sign recognition, is evaluated through several experiments. Based on these experiments, our proposed TIRuD reduces the accuracy by about while saving execution time about over the exact multiplication, averagely. Moreover, employing our proposed hierarchical approximation in various model layers outperforms the exact computation considering "AoC" that joins accuracy and computation cost in a parameter.

Paper Structure

This paper contains 14 sections, 1 equation, 7 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Proposed method
  4. Problem Statement
  5. AppSign: Our proposed Approximate Hierarchical Sign Detection Unit
  6. Proposed approximate multiplier (TIRuD)
  7. Experimental result
  8. Simulation setup
  9. Experiment
  10. Efficiency of Our Proposed Approximate Multiplier (TIRuD)
  11. Efficiency of Our Proposed Multi-level AppSign
  12. Compromising the Accuracy and Computation overhead
  13. Comparison to Related Studies
  14. Conclusion

Figures (7)

  • Figure 1: The architecture of the employed CNN and the approximation of the layers in AppSign regarding the computation costs
  • Figure 2: Samples of the traffic sign images of the employed Dataseta51
  • Figure 3: Summary of the employed CNN model for traffic sign recognition
  • Figure 4: Statistical accuracy of employing non-uniform approximation on two convolution layers of the model. (The horizontal red line shows the accuracy of the exact method.)
  • Figure 5: Accuracy of combining low, high, and high-precision approximation methods on three layers of the model (SX:Shift Xor, R: Rounding, L: LNS, F: FAMM, Q: Quantize, T: TIRuD, SA: Shift add). (The horizontal red line is the accuracy of the exact method.)
  • ...and 2 more figures