Dynamic Vision Sensor Object Detection Using Recurrent Coherent Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection methods based on convolutional neural networks require significant storage and computing resources, making them unsuitable for embedded electronic applications such as automatic driving and unmanned aerial vehicles, despite their high detection performance, as they are typically run on GPUs and do not meet the power efficiency requirements of these applications.
Innovation Solution
An object detection method utilizing a Dynamic Vision Sensor (DVS) with a novel recurrent coherent network that includes a frame detection network model and a candidate graph model, which processes image frames to generate candidate detection boxes by leveraging temporal information and reducing computational overhead through a lightweight network structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional object detection methods based on convolutional neural networks are used, then detection accuracy is improved, but power consumption and computational resource requirements increase significantly
Solution Approach 1:
The patent segments the object detection task into two distinct phases: a coarse detection phase that quickly identifies potential object regions, and a fine detection phase that accurately classifies objects within those regions. This segmentation allows the system to avoid applying computationally intensive operations to the entire image, thereby reducing overall power consumption while maintaining detection accuracy.
Solution Approach 2:
The patent applies partial action by performing full computational processing only on regions of interest identified in the coarse detection phase, rather than processing the entire image with the same computational intensity. This selective processing reduces the total computational load and power consumption while ensuring accurate detection where needed.
2Measurement precision
If conventional object detection methods based on convolutional neural networks are used, then detection accuracy is improved, but device complexity increases
Solution Approach 1:
The detection system is divided into two stages: a lightweight coarse detection stage that rapidly identifies candidate regions, and a more sophisticated fine detection stage that performs accurate classification only on those candidates. This segmentation reduces the computational complexity burden on the device by distributing tasks appropriately across stages.
Solution Approach 2:
The patent implements dynamic processing where the level of computational detail applied varies based on the detection stage and region of interest. The system dynamically adjusts processing intensity, applying simple operations in the coarse stage and more complex operations only when and where necessary in the fine stage, thereby reducing overall device complexity requirements.
3Productivity
If faster detection algorithms are used to increase detection speed, then processing time is reduced, but detection accuracy decreases
Solution Approach 1:
The patent segments detection into a fast coarse phase that prioritizes speed and a slower fine phase that prioritizes accuracy. By processing most of the image in the fast coarse phase and reserving detailed analysis only for identified regions of interest, the system achieves high overall detection speed while maintaining accuracy through the subsequent fine detection phase.
Solution Approach 2:
The system applies partial detailed processing only to regions of interest rather than performing exhaustive analysis on the entire image. This allows the majority of processing to occur at lower computational levels for speed, while still achieving high accuracy through targeted detailed analysis where needed.
Data Source
AI summary
The disclosure provides an object detection method and apparatus based on a Dynamic Vision Sensor (DVS). The method includes the following operations of: acquiring a plurality of image frames by a DVS; and, detecting the image frames by a recurrent coherent network to acquire a candidate box for objects to be detected, wherein the recurrent coherent network comprising a frame detection network model and a candidate graph model. By using a new recurrent coherent detection network, a bounding box for an object to be detected is fast detected from the data acquired by a DVS. The detection speed is improved greatly while ensuring the detection accuracy.