.png){kind=icon}
.png){kind=icon}
Return
Content-Aware Low-Complexity Object Detection for Tracking Using Adaptive Compressed Sensing
.png){kind=icon}
OA
AI
DOI：10.1109/JETCAS.2018.2859218.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Performing object detection for tracking for high resolution videos poses a challenge for real-time operation on embedded systems with limited resources. Fixed-camera background segmentation methods require creating and updating independent statistical models for each pixel in the captured scene, resulting in millions of models that must be maintained and evaluated with each captured frame. This research presents a minimal-complexity object tracking system design that modulates system throughput based on the salience of the scene. Each frame is adaptively sampled with spatially varying density, such that regions containing foreground objects are sampled with high density and background regions are sampled with very low density. Background regions are sampled just densely enough to validate that the region has not changed. A novel control system is proposed to evaluate post-processed information and continually adjust the sampling density of each frame. As a result, the computational complexity of the entire system scales with the amount of foreground object activity in the scene. With granular control of voltage and frequency scaling, system resources are scaled to just meet the requirements of the observed scene. Experiments show an average reduction of computational complexity by 50%.
Keywords:
Object detection
compressed sensing
subtraction techniques
video surveillance
adaptive systems
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
3.8
Papers:
1.4K
Citations:
2.8K