- 學位論文
惡劣天氣條件下車輛重新識別的圖像自適應
Image-Adaptive for Vehicle Re-identification in Adverse Weather Conditions
若您是本文的作者，可授權文章由華藝線上圖書館中協助推廣。
摘要
車輛重識別在電腦視覺中引起廣泛的關注。儘管現階段已經提出了幾種最先進方法來解決，但從惡劣天氣條件下拍攝的低質量影像中，重新識別車輛在霧和弱光場景仍然具有挑戰性。據我所知，這個問題仍然沒有得到很好的解决。為了緩解這個問題，我提出了一種新的影像自我調整車輛重識別框架，能讓每個影像都自我調整並增强，以獲得更好的重識別效能。 具體而言，我使用了一種可微分圖像處理模塊解決惡劣天氣條件對重識別檢測器的影響，其參數由小型卷積神經網路預測。我以端到端的方式聯合學習小型卷積網路和重識別網路，這確保了小型卷積網路可以學習適當的參數，以弱監督的方上增强影像進行重識別。由於缺乏專門用於惡劣天氣下車輛重識別的數據集，我建構了一個名為Wild_Foggy和一個名為V1M_lowlight的數據集，分別由真實世界和合成的霧影像及低光源影像組成，用於訓練和評估效能。我提出的方法可以在正常和惡劣天氣條件下自我調整地處理影像。實驗結果表明，在霧天和微光天氣下，該方法都是有效的，優於現有的其他車輛重識別方法。
並列摘要
Vehicle Re-identification (ReID) has attracted considerable attention in computer vision. Although several methods have been proposed to achieve state-of-the-art performance on this topic, it is still challenging to re-identifying vehicle foggy and low-light scenes from the low-quality images captured in adverse weather conditions. To my knowledge, this problem is still not well-addressed. To alleviate this problem, I propose a novel Image-Adaptive Re-identification (IAVE-ReID) framework, where each image can be adaptively enhanced for better Re-identification performance. Specifically, a Differentiable Image Processing (DIP) module is proposed to consider adverse weather conditions for the ReID detector, whose parameters are predicted by a small convolutional neural network (CNN-PD). I jointly learn CNN-PD and ReID in an end-to-end manner, which ensures that CNN-PD can learn proper DIP to enhance images for classification in a weakly supervised manner. Due to the lack of datasets dedicated to vehicle-ReID in adverse weather, I constructed a foggy dataset named Wild_Foggy and a dataset named V1M_lowlight which consist of real-world and synthetic foggy images and low-light image composition for training and evaluating performance. My proposed IAVE-ReID method can adaptively process images under normal and severe weather conditions. Experimental results show that the proposed method is effective and outperforms other existing vehicle ReID methods in foggy and low-light weather.
參考文獻
[1] Xinchen Liu, Wu Liu, Tao Mei, and Huadong Ma. A deep learning-based approach to progressive vehicle ReIDentification for urban surveillance. In ECCV, pages 869–884, 2016.
[2] Xinchen Liu, Wu Liu, Tao Mei, and Huadong Ma. Provid: Progressive and multimodal vehicle ReIDentification for large-scale urban surveillance. T-MM, 20(3):645–658, 2017.
[3] Lou, Y.; Bai, Y.; Liu, J.;Wang, S.; and Duan, L. 2019. Veri-wild: A large dataset and a new method for vehicle Re-identification in the wild. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 3235–3243.
[4] He, B.; Li, J.; Zhao, Y.; and Tian, Y. 2019. Part-regularized near-duplicate vehicle Re-identification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 3997–4005.
[5] Orientation-aware vehicle Re-identification with semantics-guided part attention network. In European Conference on Computer Vision, 330–346. Springer.