Panopticus: Omnidirectional 3D Object Detection on Resource-constrained Edge Devices

Jeho Lee,Chanyoung Jung,Jiwon Kim,Hojung Cha

2024-10-02

Abstract:3D object detection with omnidirectional views enables safety-critical applications such as mobile robot navigation. Such applications increasingly operate on resource-constrained edge devices, facilitating reliable processing without privacy concerns or network delays. To enable cost-effective deployment, cameras have been widely adopted as a low-cost alternative to LiDAR sensors. However, the compute-intensive workload to achieve high performance of camera-based solutions remains challenging due to the computational limitations of edge devices. In this paper, we present Panopticus, a carefully designed system for omnidirectional and camera-based 3D detection on edge devices. Panopticus employs an adaptive multi-branch detection scheme that accounts for spatial complexities. To optimize the accuracy within latency limits, Panopticus dynamically adjusts the model's architecture and operations based on available edge resources and spatial characteristics. We implemented Panopticus on three edge devices and conducted experiments across real-world environments based on the public self-driving dataset and our mobile 360° camera dataset. Experiment results showed that Panopticus improves accuracy by 62% on average given the strict latency objective of 33ms. Also, Panopticus achieves a 2.1{\times} latency reduction on average compared to baselines.

Computer Vision and Pattern Recognition,Systems and Control

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenge of achieving omnidirectional 3D object detection on resource - constrained edge devices. Specifically, the paper points out that although 3D object detection is crucial in safety - critical applications such as mobile robot navigation, current solutions often require high - performance computing devices to process large amounts of sensor data, which is difficult to achieve on edge devices. Edge devices usually have limited computing power and cannot support high - precision 3D detection tasks, especially in the case of requiring an omnidirectional view. In addition, although using low - cost cameras instead of expensive LiDAR sensors can reduce costs, how to meet strict latency requirements while ensuring detection accuracy has become an urgent problem to be solved.

To address these challenges, the paper proposes the Panopticus system. Panopticus is a carefully designed system aimed at solving the above problems through an adaptive multi - branch detection scheme that can adjust the model architecture and operations according to spatial complexity. Panopticus can improve detection accuracy while maintaining low latency and is especially suitable for resource - constrained edge devices. The system optimizes accuracy under strict latency constraints by dynamically adjusting the model architecture to adapt to the available resources and spatial characteristics of edge devices. Experimental results show that Panopticus improves the detection accuracy by 62% with an average latency target of 33 milliseconds and reduces the average latency by 2.1 times compared to the baseline method.