Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DECADE: Towards Designing Efficient-yet-Accurate Distance Estimation Modules for Collision Avoidance in Mobile Advanced Driver Assistance Systems

Muhammad Zaeem Shahzad, Muhammad Abdullah Hanif, Muhammad Shafique

TL;DR

A distance estimation model, DECADE, is presented that processes each detector output instead of constructing pixel-wise depth/disparity maps and a pose estimation DNN is proposed to estimate allocentric orientation of detections to supplement the distance estimation DNN in its prediction of distance using bounding box features.

Abstract

The proliferation of smartphones and other mobile devices provides a unique opportunity to make Advanced Driver Assistance Systems (ADAS) accessible to everyone in the form of an application empowered by low-cost Machine/Deep Learning (ML/DL) models to enhance road safety. For the critical feature of Collision Avoidance in Mobile ADAS, lightweight Deep Neural Networks (DNN) for object detection exist, but conventional pixel-wise depth/distance estimation DNNs are vastly more computationally expensive making them unsuitable for a real-time application on resource-constrained devices. In this paper, we present a distance estimation model, DECADE, that processes each detector output instead of constructing pixel-wise depth/disparity maps. In it, we propose a pose estimation DNN to estimate allocentric orientation of detections to supplement the distance estimation DNN in its prediction of distance using bounding box features. We demonstrate that these modules can be attached to any detector to extend object detection with fast distance estimation. Evaluation of the proposed modules with attachment to and fine-tuning on the outputs of the YOLO object detector on the KITTI 3D Object Detection dataset achieves state-of-the-art performance with 1.38 meters in Mean Absolute Error and 7.3% in Mean Relative Error in the distance range of 0-150 meters. Our extensive evaluation scheme not only evaluates class-wise performance, but also evaluates range-wise accuracy especially in the critical range of 0-70m.

DECADE: Towards Designing Efficient-yet-Accurate Distance Estimation Modules for Collision Avoidance in Mobile Advanced Driver Assistance Systems

TL;DR

A distance estimation model, DECADE, is presented that processes each detector output instead of constructing pixel-wise depth/disparity maps and a pose estimation DNN is proposed to estimate allocentric orientation of detections to supplement the distance estimation DNN in its prediction of distance using bounding box features.

Abstract

The proliferation of smartphones and other mobile devices provides a unique opportunity to make Advanced Driver Assistance Systems (ADAS) accessible to everyone in the form of an application empowered by low-cost Machine/Deep Learning (ML/DL) models to enhance road safety. For the critical feature of Collision Avoidance in Mobile ADAS, lightweight Deep Neural Networks (DNN) for object detection exist, but conventional pixel-wise depth/distance estimation DNNs are vastly more computationally expensive making them unsuitable for a real-time application on resource-constrained devices. In this paper, we present a distance estimation model, DECADE, that processes each detector output instead of constructing pixel-wise depth/disparity maps. In it, we propose a pose estimation DNN to estimate allocentric orientation of detections to supplement the distance estimation DNN in its prediction of distance using bounding box features. We demonstrate that these modules can be attached to any detector to extend object detection with fast distance estimation. Evaluation of the proposed modules with attachment to and fine-tuning on the outputs of the YOLO object detector on the KITTI 3D Object Detection dataset achieves state-of-the-art performance with 1.38 meters in Mean Absolute Error and 7.3% in Mean Relative Error in the distance range of 0-150 meters. Our extensive evaluation scheme not only evaluates class-wise performance, but also evaluates range-wise accuracy especially in the critical range of 0-70m.

Paper Structure

This paper contains 14 sections, 2 equations, 17 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. Motivational Case Study
  3. Our Novel Contributions
  4. Methodology
  5. Feature Analysis
  6. DNNs in DECADE
  7. Adaptation to Detector Error
  8. Evaluation and Discussion
  9. Experimental Setup
  10. Evaluation on Ground Truth
  11. End-to-End Evaluation with YOLOv8
  12. Adapted Networks
  13. Evaluating the Accuracy-Efficiency Tradeoff
  14. Conclusion

Figures (17)

  • Figure 1: Overview of Our Mobile Advanced Driver-Assistance System (ADAS)
  • Figure 2: Pixel-wise vs Detection-wise Distance Estimation. (a) presents pixel-wise distance estimation where the input image is processed without detector involvement. (b) presents detection-wise distance estimation where only the features extracted from the detector's outputs are processed.
  • Figure 3: Distance estimation accuracy vs. parametric complexity of state-of-the-art collision avoidance DL-implementation modules for mobile ADAS. The labeled distance estimation models are combined with the YOLOv8 n/s/m object detection variants for a fair comparison across pixel-wise and detection-wise domains, except for the Dist-YOLO in which distance estimation is embedded into the YOLOv3 architecture.
  • Figure 4: Overview of the DECADE model, attached to an object detection network
  • Figure 5: Relationship between dimensional features (normalized by image dimensions) of bounding boxes and the object's distance from the camera
  • ...and 12 more figures