Search

In recent years, automotive radar has become an integral part of the advanced safety sensor stack. Although radar gives a significant advantage over a camera or Lidar, it suffers from poor angular resolution, unwanted noises and significant object smearing across the angular bins, making radar-based object detection challenging. We propose a novel radar-based object detection utilizing a deep learning-based super-resolution (DLSR) model.

This paper proposes a multichannel method for discriminative region localization in Camouflaged Object Detection (COD) tasks. In one channel, processing the phase and amplitude of 2-D Fourier spectra generate a modified form of the original image, used later for a pixel-wise optimal local entropy analysis. The other channel implements a class activation map (CAM) and Global Average Pooling (GAP) for object localization.

In recent years, saliency object detection methods based on convolutional neural networks have been widely studied,and have achieved excellent performance in clear images. However, due to the low visibility of images in foggy conditions, the existing saliency object detection methods will be seriously affected or even ineffective. To address this problem, we introduce an end-to-end multi-task learning network.

Wide area urban street monitoring is highly demanding in various smart city applications. Manual monitoring is both laborious and time-consuming, hence automatic vision-based monitoring is a more feasible alternative. An essential part of vision-based street monitoring is detecting and counting objects of interest. However, these tasks are not straightforward due to various challenges, i.e., noisy conditions in a real environment, such as occlusion and high illumination.

Deep learning advances have propelled common object detection development. However, small object detection in aerial images remains inaccurate. Intersection-over-Union (IoU) has limitations in handling the separation or inclusion cases of prediction and ground truth boxes, especially in small object detection, which significantly hampers the process of bounding box regression.

Bounding boxes are predominant annotation and detection methods in object detection, and their accuracies are commonly gauged using the intersection-over-union (IoU) index and its derivatives. While these methods are effective for clearly defined objects, modifications are required when using bounding box classifiers with ambiguously defined objects.

Excellent results can be achieved using object detection with fully supervised training on large well-annotated datasets. However, the problem of missing annotations in real-world datasets can considerably reduce the performance of object detectors. In this study, we thoroughly analyze the effect of missing annotations on both positive and negative samples in object detector training.

Salient object detection (SOD) is a task that involves identifying and segmenting the most visually prominent object in an image. Existing solutions can accomplish this using a multi-scale feature fusion mechanism to detect the global context of an image. However, as there is no consideration of the structures in the image nor the relations between distant pixels, conventional methods cannot deal with complex scenes effectively.

Cross-modal information fusion plays a vital role in the RGB-T salient object detection. Due to RGB and thermal images come from different domains, the modality difference will lead to the unsatisfactory effect of simple feature fusion. How to explore and integrate useful information is the key to the RGB-T saliency detection methods. In this paper, we introduce an Enhancement and Fusion Network.

A new research topic, multi-source domain adaptive object detection (MSDAOD) under privacy-preserving constraint is explored in this paper, where the clients can only access their own source data while the server can only access unlabeled target data.