A comparison of CNN-based face and head detectors for real-time video surveillance applications

Nguyen-Meidine, Le Thanh; Granger, Éric; Kiran, Madhu; Blais-Morin, Louis-Antoine

doi:10.1109/ipta.2017.8310113

Cited by 28 publications

(12 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The goal of target detection is to determine the spatial location and the category of the targets. At present, target detection has been widely used in our daily life [1], [2], such as pedestrian detection [3], face detection [4], [5], vehicle detection [6]- [8], intelligent surveillance [9]- [11] and autonomous driving [12], [13], etc.…”

Section: Introductionmentioning

confidence: 99%

A Simple and Efficient Network for Small Target Detection

Luo

Zhang

et al. 2019

IEEE Access

View full text Add to dashboard Cite

Target detection based on deep learning is developing rapidly. However, small target detection is still a challenge. In this paper, a simple and efficient network for small target detection is proposed. We put forward to improve the detection performance of the small targets in three aspects. First, as the contextual information is important to detect the small targets, we proposed to use ''dilated module'' to expand the receptive field without loss of resolution or coverage. Second, we applied feature fusion in different dilated modules to improve the ability of the network in detecting small targets. Finally, we used ''passthrough module'' to get the finer-grained information from the earlier layer and combined it with the semantic information from the deeper layer. To improve the detection speed of the network, it is proposed to use 1 × 1 convolution to reduce the dimension of the network. We composed small vehicle dataset based on VEDAI dataset and DOTA dataset, respectively, and also analyzed the distribution of the small targets in each dataset. To evaluate the performance of the proposed network, we trained the model on the dataset above and compared with the state-of-the-art target detection algorithms, our approach achieved 80.16% average precision (AP) on VEDAI dataset and 88.63% AP on DOTA dataset and the frames per second (FPS) is 75.4. The AP of our network is much better than the result of the tiny YOLO V3 and is nearly the same as the result of the YOLO V3. However, the FPS of our network is almost the same as that of the tiny YOLO V3.

show abstract

Section: Introductionmentioning

confidence: 99%

A Simple and Efficient Network for Small Target Detection

Luo

Zhang

et al. 2019

IEEE Access

View full text Add to dashboard Cite

show abstract

“…The residual connections are important in deep architectures. They significantly accelerate the training of the inception network ( 27 , 28 ) and provide an efficient means of dealing with the degradation problem caused by the deep network ( 7 ). Figure 5 shows the overall architecture of Faster R-CNN.…”

Section: Methodsmentioning

confidence: 99%

A Scalable Deep Learning Framework for Extracting Model Inventory of Roadway Element Intersection Control Types From Panoramic Images

Kwayu

Toth

Himmelein

2022

Transportation Research Record: Journal of the Transportation R

View full text Add to dashboard Cite

In the United States, the Model Inventory of Roadway Element (MIRE) provides a comprehensive list of data that are needed to support states’ data-driven safety programs. The intersection control is part of the MIRE Fundamental Data Elements (FDE) for which state Departments of Transportation are required to complete the collection by September 30, 2026. It is essential roadway data that have been used widely in traffic safety studies. This study proposes a scalable and automated deep learning framework for detecting and classifying stop and yield intersection controls using panoramic street view images. The Faster Region-based Convolutional Neural Networks (Faster R-CNN) model architecture was used to detect and classify stop and yield signs from the images. A transfer learning process was deployed using the Inception-ResNet-v2 generic feature extractor to accelerate the training and performance of the deep learning model with less data collection effort. The effectiveness and scalability of the proposed framework were tested on a sample of road intersections in the state of Michigan. The proposed deep learning model achieved a recall value of 97.7% and 98.2% for detecting and classifying stop and yield signs respectively. The evaluation of the model performance at a county level suggests that the model can be scaled to a statewide level without a substantial increase in the demand for computational resources. As demonstrated in this study, state DOTs can leverage the advancement of deep learning techniques and the availability of imagery data to expedite the process of collecting the MIRE data.

show abstract

“…These push the evaluation of such networks for realtime semantic segmentation or object detection out of reach of even the most powerful embedded platforms available today for high-resolution video data [14]. However, exactly such systems are required for a wide range of applications limited in cost (CCTV/urban surveillance, perimeter surveillance, consumer behavior and highway monitoring) and latency (aerospace and UAV monitoring and defense, visual authentication) [15], [16].…”

Section: Introductionmentioning

confidence: 99%

CBinfer: Exploiting Frame-to-Frame Locality for Faster Convolutional Network Inference on Video Streams

Cavigelli

Benini

2020

IEEE Trans. Circuits Syst. Video Technol.

View full text Add to dashboard Cite

The last few years have brought advances in computer vision at an amazing pace, grounded on new findings in deep neural network construction and training as well as the availability of large labeled datasets. Applying these networks to images demands a high computational effort and pushes the use of state-of-the-art networks on real-time video data out of reach of embedded platforms.Many recent works focus on reducing network complexity for real-time inference on embedded computing platforms. We adopt an orthogonal viewpoint and propose a novel algorithm exploiting the spatio-temporal sparsity of pixel changes. This optimized inference procedure resulted in an average speed-up of 9.1× over cuDNN on the Tegra X2 platform at a negligible accuracy loss of < 0.1% and no retraining of the network for a semantic segmentation application. Similarly, an average speed-up of 7.0× has been achieved for a pose detection DNN and a reduction of 5× of the number of arithmetic operations to be performed for object detection on static camera video surveillance data. These throughput gains combined with a lower power consumption result in an energy efficiency of 511 GOp/s/W compared to 70 GOp/s/W for the baseline.

show abstract

A comparison of CNN-based face and head detectors for real-time video surveillance applications

Cited by 28 publications

References 11 publications

A Simple and Efficient Network for Small Target Detection

A Simple and Efficient Network for Small Target Detection

A Scalable Deep Learning Framework for Extracting Model Inventory of Roadway Element Intersection Control Types From Panoramic Images

CBinfer: Exploiting Frame-to-Frame Locality for Faster Convolutional Network Inference on Video Streams

Contact Info

Product

Resources

About