基於空間正規化區域生成網路之無人機物體計數

Drone-based Object Counting by Spatially Regularized Regional Proposal Networks

謝孟儒(Meng-Ru Hsieh)

指導教授 ： 徐宏民

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2017年)

https://doi.org/10.6342/NTU201704217

摘要

現有的計數方式時常採用迴歸式的方法，且沒辦法精準定位目標物體位置所在，使得更進一步的應用和分析被限制，例如更高層次的理解和強化辨識。除此之外，大多數先前的研究著重於使用已架好的攝影機在固定的場景下計數。隨著無人駕駛機的出現，例如drone，我們對動態環境下的物體進行定位和計數產生了濃厚的興趣。我們提出了布局生成網路和空間核，在drone的影片中來同時計數及定位目標物體，例如車子。不同於傳統的區域生成網路，我們引進了空間布局訊息，例如車子通常在停車場會規律的停放，和利用了空間正規化限制到我們的網路，來改善定位精準度。為了評估我們的計數方法，我們也提供了最新大規模停車場資料集(CARPK)，裡面含有從不同停車場拍攝的將近九萬台車輛。就我們所知，這個是目前世界上第一個、同時也是最大能支援物體計數的無人機視角資料集，且提供邊框標註。

我們的貢獻包括:
1.目前所知, 這是第一個有在區域生成網路中利用空間布局訊息的研究。在公開的PKLot資料集的子資料集PUCPR上，我們改善了目前最好區域生成網路方法的了平均召回率，例如從59.9%提生至62.5%。

2.我們提供了最新大規模的停車場資料集(CARPK)，裡面含有9萬張從不同停車場場景拍攝的無人機視角高解析度影像。最重要的是，跟其他停車場資料集相比，我們的CARPK資料集是世界上第一個、同時也是最大能支援物體計數的無人機視角資料集。

3.我們提供對於區域生成網路不同決策選擇的深入分析，並展示利用布局訊息可以很可觀的減少生成數量同時改善計數結果。

關鍵字

計數 ； 物體生成 ； 電腦視覺

並列摘要

Existing counting methods often adopt regression-based approaches and cannot precisely localize the target objects, which hinders the further analysis (e.g., high-level understanding and fine-grained classification). In addition, most of prior work mainly focus on counting objects in static environments with fixed cameras. Motivated by the advent of unmanned flying vehicles (i.e., drones), we are interested in detecting and counting objects in such dynamic environments. We propose Layout Proposal Networks (LPNs) and spatial kernels to simultaneously count and localize target objects (e.g., cars) in videos recorded by the drone. Different from the conventional region proposal methods, we leverage the spatial layout information (e.g., cars often park regularly) and introduce these spatially regularized constraints into our network to improve the localization accuracy. To evaluate our counting method, we present a new large-scale car parking lot dataset (CARPK) that contains nearly 90,000 cars captured from different parking lots. To the best of our knowledge, it is the first and the largest drone view dataset that supports object counting, and provides the bounding box annotations.

Our contributions include:
1.To our knowledge, this is the first work that leverages spatial layout information for object region proposal. We improve the average recall of the state-of-the-art region proposal methods (i.e., 59.9% to 62.5%) on a public PUCPR dataset.

2.We introduce a new large-scale car parking lot dataset (CARPK) that contains nearly 90,000 cars in drone-based high resolution images recorded from the diverse scenes of parking lots. Most important of all, compared to other parking lot datasets, our CARPK dataset is the first and the largest dataset of parking lots that can support counting.

3.We provide in-depth analyses for different decision choices of our region proposal method, and demonstrate that utilizing layout information can considerably reduce the proposals and improve the counting results.

並列關鍵字

Counting ； Object Proposal ； Computer Vision

參考文獻

[1] M. Ahrnbom, K. Astrom, and M. Nilsson. Fast classification of empty and occupied parking spaces using integral channel features. In CVPR, 2016.

[2] G. Amato, F. Carrara, F. Falchi, C. Gennaro, and C. Vairo. Car parking occupancy detection using smart camera networks and deep learning. In ISCC, 2016.

[3] S. An, W. Liu, and S. Venkatesh. Face recognition using kernel ridge regression. In CVPR, 2007.

[4] P. Arbel´aez, J. Pont-Tuset, J. T. Barron, F. Marques, and J. Malik. Multiscale combinatorial grouping. In CVPR, 2014.

[5] C. Arteta, V. Lempitsky, J. A. Noble, and A. Zisserman. Interactive object counting. In ECCV, 2014.