Multi-View Object Detection Based on Deep Learning

This paper aims at tackling the task of fusion feature from images and their corresponding point clouds for 3D object detection in autonomous driving scenarios based on AVOD, an Aggregate View Object Detection network. The proposed fusion algorithms fuse features targeted from Bird’s Eye View (BEV) LIDAR point clouds and their corresponding RGB images. Differing in existing fusion methods, which are simply the adoption of the concatenation module, the element-wise sum module or the element-wise mean module, our proposed fusion algorithms enhance the interaction between BEV feature maps and their corresponding image feature maps by designing a novel structure, where single level feature maps and utilize multilevel feature maps. Experiments show that our proposed fusion algorithm produces better results on 3D mAP and AHS with less speed loss compared to the existing fusion method used on the KITTI 3D object detection benchmark.

“…Ref. [7] utilizes RGB images and depth images with early fusion strategy and trains pose-based classifiers for 2D detection [14]. Refs.…”

Section: Related Work and Proposed Methodsmentioning

confidence: 99%

A Novel Interactive Fusion Method with Images and Point Clouds for 3D Object Detection

Yang

et al. 2019

“…For example, Cha, et al [36] used 256 × 256 small images to participate in training after image pre-processing and then detected 5888 × 3584 large images based on convolutional neural networks. Tang, et al [37] proposed a multi-view object detection method based on deep learning, due to the weak ability of detecting small objects in classical object detection methods based on regression models, and experimented with multi-view YOLO, YOLO2, SSD, improving the accuracy and speed in small object detection; Tayara, et al [38] proposed object detection in very high-resolution aerial images using a one-stage densely connected feature pyramid network, by which high-level multi-scale semantic feature maps with high-quality information are prepared for object detection. This work has been evaluated on two publicly available datasets and outperformed the current state-of-the-art results both in terms of mean average precision (mAP) and computation time.…”

Section: Introductionmentioning

confidence: 99%

Object Detection and Classification of Metal Polishing Shaft Surface Defects Based on Convolutional Neural Network Deep Learning

Qingsheng

Tan

et al. 2019

Defective shafts need to be classified because some defective shafts can be reworked to avoid replacement costs. Therefore, the detection and classification of shaft surface defects has important engineering application value. However, in the factory, shaft surface defect inspection and classification are done manually, with low efficiency and reliability. In this paper, a deep learning method based on convolutional neural network feature extraction is used to realize the object detection and classification of metal shaft surface defects. Through image segmentation, the system methods setting of a Fast-R-CNN object detection framework and parameter optimization settings are implemented to realize the classification of 16,384 × 4096 large image little objects. The experiment proves that the method can be applied in practical production and can also be extended to other fields of large image micro-fine defects with a high light surface. In addition, this paper proposes a method to increase the proportion of positive samples by multiple settings of IOU values and discusses the limitations of the system for defect detection.

“…Tang et al proposed a multi-view YOLO object detection method, model structure and working method, which improved the ability to detect small objects. Compared with the two-stage method, this multi-view one-stage method is faster when it reaches the same map when implementing small object detection [2].…”

Section: Introductionmentioning

confidence: 99%

Application Research of Improved YOLO V3 Algorithm in PCB Electronic Component Detection

Huang

et al. 2019

110

Target detection of electronic components on PCB (Printed circuit board) based on vision is the core technology for 3C (Computer, Communication and Consumer Electronics) manufacturing companies to achieve quality control and intelligent assembly of robots. However, the number of electronic components on PCB is large, and the shape is different. At present, the accuracy of the algorithm for detecting all electronic components is not high. This paper proposes an improved algorithm based on YOLO (you only look once) V3 (Version 3), which uses a real PCB picture and a virtual PCB picture with synthesized data as a joint training dataset, which greatly increases the recognizability of training electronic components and provides the greatest possibility for data enhancement. After analyzing the feature distribution of the five dimensionality-reduced output layers of Darknet-53 and the size distribution of the detection target, it is proposed to adjust the original three YOLO output layers to four YOLO output layers and generate 12 anchor boxes for electronic component detection. The experimental results show that the mean average precision (mAP) of the improved YOLO V3 algorithm can achieve 93.07%.