Understanding How Video Quality Affects Object Detection Algorithms

Aqqa, Miloud; Mantini, Pranav; Shah, Shishir K.

doi:10.5220/0007401600960104

Cited by 18 publications

(10 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Similarly, the performance of object detection (e.g., pedestrian, car, traffic sign, etc.) algorithms heavily depends on the image quality [ 6 , 7 ]. As a consequence, monitoring image/video quality is also crucial in vision-based advanced driver-assistance systems [ 8 ].…”

Section: Introductionmentioning

confidence: 99%

No-Reference Image Quality Assessment Based on the Fusion of Statistical and Perceptual Features

Varga

2020

J. Imaging

View full text Add to dashboard Cite

The goal of no-reference image quality assessment (NR-IQA) is to predict the quality of an image as perceived by human observers without using any pristine, reference images. In this study, an NR-IQA algorithm is proposed which is driven by a novel feature vector containing statistical and perceptual features. Different from other methods, normalized local fractal dimension distribution and normalized first digit distributions in the wavelet and spatial domains are incorporated into the statistical features. Moreover, powerful perceptual features, such as colorfulness, dark channel feature, entropy, and mean of phase congruency image, are also incorporated to the proposed model. Experimental results on five large publicly available databases (KADID-10k, ESPL-LIVE HDR, CSIQ, TID2013, and TID2008) show that the proposed method is able to outperform other state-of-the-art methods.

show abstract

Section: Introductionmentioning

confidence: 99%

No-Reference Image Quality Assessment Based on the Fusion of Statistical and Perceptual Features

Varga

2020

J. Imaging

View full text Add to dashboard Cite

show abstract

“…The roof, unknown objects, and the bus are error-detected objects. Due to the low-quality video (Aqqa et al, 2019) and small dataset (Cao et al, 2019) used to train Faster RCNN, it cannot detect vehicles as accurately as YOLO in Table 4. According to Cao et al (2019), the researchers proposed an improved Faster RCNN and used the TT100K (Tsinghua-Tencent 100K) dataset, which saves 100,000 images, including 30,000 traffic-sign occurrences.…”

Section: Faster Rcnn Resultsmentioning

confidence: 99%

“…Figure 16 shows the detection of errors in high-quality and low-quality videos. According to Aqqa et al (2019), video quality is an important factor, often overlooked. The video is tested using Faster RCNN, SSD, YOLO, and RetinaNet for object detection at various video compression levels to investigate the quality distortion caused by compression artefacts during video capture.…”

Section: Faster Rcnn Resultsmentioning

confidence: 99%

Data Safety Prediction Using Bird’s Eye View and Social Distancing Monitoring for Penang Roads

Ming¹,

Ali²,

Ismail³

et al. 2022

JST

View full text Add to dashboard Cite

In terms of fatalities, Malaysia ranks third among ASEAN countries. Every year, there is an increase in accidents and fatalities. The state of the road is one factor contributing to near misses. A near miss is an almost-caused accident, an unplanned situation that could result in injury or accidents. The Majlis Bandar Pulau Pinang (MBPP) has installed 1841 closed-circuit television (CCTV) cameras around Penang to monitor traffic and track near miss incidents. When installing CCTVs, the utilisation of video allows resources to be used and optimised in situations when maintaining video memories is difficult and costly. Highways, industrial regions, and city roads are the most typical places where accidents occur. Accidents occurred at 200 per year on average in Penang from 2015 to 2017. Near misses are what create accidents. One of the essential factors in vehicle detection is the “near miss.” In this study, You Only Look Once version 3 (YOLOv3) and Faster Region-based Convolutional Neural Network (Faster RCNN) are used to solve transportation issues. In vehicle detection, a faster RCNN was used. Bird’s Eye View and Social Distancing Monitoring are used to detect the only vehicle in image processing and observe how near misses occur. This experiment tests different video quality and lengths to compare test time and error detection percentage. In conclusion, YOLOv3 outperforms Faster RCNN. In high-resolution videos, Faster RCNN outperforms YOLOv3, while in low-resolution videos, YOLOv3 outperforms Faster RCNN.

show abstract

“…Streaming's impact on DNN. Some measurement studies have been conducted to understand how the quality of the video itself, e.g., resolution, affects the accuracy of deep object detection [24], [25]. For example, Dodge and Karam [25] figured out that the inference accuracy is susceptible to blur and noise distortions while being resilient to compression artifacts and contrast.…”

Section: Related Workmentioning

confidence: 99%

Balancing Latency and Accuracy on Deep Video Analytics at the Edge

Cho

Xiao

2022

2022 IEEE 19th Annual Consumer Communications &Amp; Networking Conference (CCNC)

View full text Add to dashboard Cite

Real-time deep video analytic at the edge is an enabling technology for emerging applications, such as vulnerable road user detection for autonomous driving, which requires highly accurate results of model inference within a low latency. In this paper, we investigate the accuracy-latency trade-off in the design and implementation of real-time deep video analytic at the edge. Without loss of generality, we select the widely used YOLObased object detection and WebRTC-based video streaming for case study. Here, the latency consists of both networking latency caused by video streaming and the processing latency for video encoding/decoding and model inference. We conduct extensive measurements to figure out how the dynamically changing settings of video streaming affect the achieved latency, the quality of video, and further the accuracy of model inference. Based on the findings, we propose a mechanism for adapting video streaming settings (i.e. bitrate, resolution) online to optimize the accuracy of video analytic within latency constraints. The mechanism has proved, through a simulated setup, to be efficient in searching the optimal settings.

show abstract

Understanding How Video Quality Affects Object Detection Algorithms

Cited by 18 publications

References 0 publications

No-Reference Image Quality Assessment Based on the Fusion of Statistical and Perceptual Features

No-Reference Image Quality Assessment Based on the Fusion of Statistical and Perceptual Features

Data Safety Prediction Using Bird’s Eye View and Social Distancing Monitoring for Penang Roads

Balancing Latency and Accuracy on Deep Video Analytics at the Edge

Contact Info

Product

Resources

About