車輛密度估計--Understanding Traffic Density from Large-Scale Web Camera Data
Understanding Traffic Density from Large-Scale Web Camera Data
CVPR2017
https://arxiv.org/abs/1703.05868
本文介紹了兩個演算法用於車輛密度估計：1）OPT-RC 根據背景差得到車輛運動區域，對於影象的不同區域學習到一個對應的權值矩陣用於估計車輛密度
2）FCN-MT 使用 FCN 分割框架來進行車輛密度估計
車輛密度估計問題還是比較難的，類似於人群密度估計
Optimization Based Vehicle Density Estimation with RankConstraint(OPT-RC)
we propose a regression model to learn different weights for different blocks to increase the degrees of freedom on the weights, and embed geometry information
用一個迴歸模型來學習影象區域對應不同的密度估計權值矩陣，嵌入了幾何資訊
FCN Based Multi-Task Learning for Vehicle Counting (FCN-MT)
網路分為 convolution network, decovolution network , 將卷積層各個層的特徵融合起來，輸入到反捲積網路中進行特徵圖放大
the large buses/trucks (oversized vehicles) in close view induce sporadically large errors in the counting results. To solve this problem, we propose a deep multi-task learning framework based on FCN to jointly learn vehicle density map and vehicle count.
為了解決個別大型車輛在影象中佔有大面積導致車輛數估計有大的錯誤，這裡使用了多目標學習
- Experiments
這裡我們建立了一個數據庫 WebCamT
