ICanC: Improving Camera-based Object Detection and Energy Consumption in Low-Illumination Environments

TL;DR

ICanC targets reliable object detection in low-illumination autonomous driving while reducing headlight energy use. It integrates LiDAR and camera processing across three ROS nodes—Obstacle Detector, Danger Detector, and Light Controller—to detect obstacles, assess threats, and selectively enable headlights. The system demonstrates robust performance in physical and simulated tests, achieving solid recall and low latency with notable energy savings under selective lighting. By open-sourcing the approach and aligning with the Vehicle Programming Interface concept, ICanC contributes to sustainable AV operation and highlights directions for enhancing LiDAR-based perception in challenging conditions.

Abstract

This paper introduces ICanC (pronounced "I Can See"), a novel system designed to enhance object detection and optimize energy efficiency in autonomous vehicles (AVs) operating in low-illumination environments. By leveraging the complementary capabilities of LiDAR and camera sensors, ICanC improves detection accuracy under conditions where camera performance typically declines, while significantly reducing unnecessary headlight usage. This approach aligns with the broader objective of promoting sustainable transportation. ICanC comprises three primary nodes: the Obstacle Detector, which processes LiDAR point cloud data to fit bounding boxes onto detected objects and estimate their position, velocity, and orientation; the Danger Detector, which evaluates potential threats using the information provided by the Obstacle Detector; and the Light Controller, which dynamically activates headlights to enhance camera visibility solely when a threat is detected. Experiments conducted in physical and simulated environments demonstrate ICanC's robust performance, even in the presence of significant noise interference. The system consistently achieves high accuracy in camera-based object detection when headlights are engaged, while significantly reducing overall headlight energy consumption. These results position ICanC as a promising advancement in autonomous vehicle research, achieving a balance between energy efficiency and reliable object detection.

Figures (11)

• Figure 1: ICanC system overview
• Figure 2: Initial Danger Detector design
• Figure 3: Current Danger Detector design
• Figure 4: State diagram for the Light Controller
• Figure 5: Zebra: A general use indoor robot