A Real-time Scalable Object Detection System using Low-power HOG Accelerator VLSI

Takagi, Kenta; Tanaka, Kenji; Izumi, Shintaro; Kawaguchi, Hiroshi; Yoshimoto, Masahiko

doi:10.1007/s11265-014-0870-7

Cited by 13 publications

(10 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Both accelerators process full HD videos at 30fps in real-time. The architecture in [11] has two classification engines similar to this work, but it does not have on-chip pyramid generation to support multi-scale and does not detect deformable parts. The two classifiers in [11] can be used to detect two different object classes at a single scale, or support multi-scale detection by processing a 5-level pyramid generated and packed off-chip into two full HD frames.…”

Section: B Evaluation Resultsmentioning

confidence: 99%

“…The architecture in [11] has two classification engines similar to this work, but it does not have on-chip pyramid generation to support multi-scale and does not detect deformable parts. The two classifiers in [11] can be used to detect two different object classes at a single scale, or support multi-scale detection by processing a 5-level pyramid generated and packed off-chip into two full HD frames. Our proposed architecture supports multi-scale generation and detection on-chip and detects deformable parts, which both increase the detection accuracy and robustness, while consuming 30% less energy/pixel compared to [11].…”

Section: B Evaluation Resultsmentioning

confidence: 99%

“…The two classifiers in [11] can be used to detect two different object classes at a single scale, or support multi-scale detection by processing a 5-level pyramid generated and packed off-chip into two full HD frames. Our proposed architecture supports multi-scale generation and detection on-chip and detects deformable parts, which both increase the detection accuracy and robustness, while consuming 30% less energy/pixel compared to [11]. Table IV shows a comparison between the proposed architecture and our software implementation of the same DPM detection algorithm mapped on ODROID-XU3 board, which is an embedded platform designed for low power consumption.…”

Section: B Evaluation Resultsmentioning

confidence: 99%

See 2 more Smart Citations

A 58.6 mW 30 Frames/s Real-Time Programmable Multiobject Detection Accelerator With Deformable Parts Models on Full HD $1920\times 1080$ Videos

Suleiman

Zhang

Sze

2017

IEEE J. Solid-State Circuits

View full text Add to dashboard Cite

Abstract-This paper presents a programmable, energyefficient and real-time object detection hardware accelerator for low power and high throughput applications using deformable parts models, with 2× higher detection accuracy than traditional rigid body models. Three methods are used to address the high computational complexity of 8 deformable parts detection: classification pruning for 33× fewer part classification, vector quantization for 15× memory size reduction, and feature basis projection for 2× reduction in the cost of each classification. The chip was fabricated in a 65nm CMOS technology, and can process full high definition 1920×1080 videos at 60fps without any off-chip storage. The chip has two programmable classification engines for multi-object detection. At 30fps, the chip consumes only 58.6mW (0.94 nJ/pixel, 1168 GOPS/W). At a higher throughput of 60fps, the classification engines can be time multiplexed to detect even more than two object classes. This proposed accelerator enables object detection to be as energyefficient as video compression, which is found in most cameras today.

show abstract

Section: B Evaluation Resultsmentioning

confidence: 99%

Section: B Evaluation Resultsmentioning

confidence: 99%

Section: B Evaluation Resultsmentioning

confidence: 99%

See 1 more Smart Citation

A 58.6 mW 30 Frames/s Real-Time Programmable Multiobject Detection Accelerator With Deformable Parts Models on Full HD $1920\times 1080$ Videos

Suleiman

Zhang

Sze

2017

IEEE J. Solid-State Circuits

View full text Add to dashboard Cite

show abstract

“…6 also shows the chip output with multi-object detection (cars and pedestrians). Comparing to the detector accelerator in [6], our chip boosts the detection accuracy with multi-scale and detecting 8 deformable parts per object while consuming 30% less energy per pixel.…”

Section: Introductionmentioning

confidence: 99%

A 58.6mW real-time programmable object detector with multi-scale multi-object support using deformable parts model on 1920×1080 video at 30fps

Suleiman

Zhang

Sze

2016

2016 IEEE Symposium on VLSI Circuits (VLSI-Circuits)

View full text Add to dashboard Cite

This paper presents a programmable, energy-efficient and realtime object detection accelerator using deformable parts models (DPM), with 2x higher accuracy than traditional rigid body models. With 8 deformable parts detection, three methods are used to address the high computational complexity: classification pruning for 33x fewer parts classification, vector quantization for 15x memory size reduction, and feature basis projection for 2x reduction of the cost of each classification. The chip is implemented in 65nm CMOS technology, and can process HD (1920x1080) images at 30fps without any off-chip storage while consuming only 58.6mW (0.94nJ/pixel, 1168 GOPS/W). The chip has two classification engines to simultaneously detect two different classes of objects. With a tested high throughput of 60fps, the classification engines can be time multiplexed to detect even more than two object classes. It is energy scalable by changing the pruning factor or disabling the parts classification. Keywords: DPM, object detection, basis projection, pruning. Introduction Object detection is critical to many embedded applications that require low power and real-time processing. For example, low latency and HD images are important for autonomous control to react quickly to fast approaching objects, while low energy consumption is essential due to battery and heat limitations. Object detection involves not only classification/recognition, but also localization, which is achieved by sliding a window of a pretrained model over an image. For multi-scale detection, the window slides over an image pyramid (multiple downscaled copies of the image). Multi-scale detection is very challenging as the image pyramid results in a data expansion, which can be more than a 100x in HD images. The high computational complexity of object detection processing necessitates fast hardware implementations [1] to enable real-time processing.This paper presents a complete object detection accelerator using DPM [2] with a root and 8 parts model as shown in Fig. 1. DPM results in double the detection accuracy compared to rigid template (root only) detection. The 8 parts account for deformation such that a single model can detect objects at different poses ( Fig. 6) and increase detection confidence. However, this accuracy comes with a classification overhead of 35x more multiplications (i.e. DPM classification consumes 80% of a single detector power), making multi-object detection a challenge. A software-based DPM object detector is described in [3], which enables detection for 500x500 images at 30fps but requires a powerful fully loaded Xeon 6-core processor and 32GB of memory. In this work, the classification overhead is significantly reduced by two main techniques: Classification pruning with vector quantization (VQ) for selective part processing.  Feature basis projection for sparse multiplications.Architecture Overview Fig. 2 shows the block diagram of our detector architecture, including histogram of oriented gradients (HOG) feature pyramid gen...

show abstract

“…A widely-used technique is the integral image method [18], which provides the possibility of obtaining area-based descriptor values in a constant time. Hardware implementations, such as the feature extraction accelerator VLSI [19], can be used in portable, on-board vehicle systems or sensor networks; hardware accelerators can also increase the efficiency of computer-based solutions. However, not all algorithms can be efficiently implemented in hardware, therefore the researchers are searching for pipeline-friendly methods, which can be embedded into the vision chips [20].…”

Section: Introductionmentioning

confidence: 99%

Histogram of Oriented Gradients with Cell Average Brightness for Human Detection

Wójcikowski¹

2016

Metrology and Measurement Systems

View full text Add to dashboard Cite

A modification of the descriptor in a human detector using Histogram of Oriented Gradients (HOG) and Support Vector Machine (SVM) is presented. The proposed modification requires inserting the values of average cell brightness resulting in the increase of the descriptor length from 3780 to 3908 values, but it is easy to compute and instantly gives ≈25% improvement of the miss rate at 10 -4 False Positives Per Window (FPPW). The modification has been tested on two versions of HOG-based descriptors: the classic Dalal-Triggs and the modified one, where, instead of spatial Gaussian masks for blocks, an additional central cell has been used. The proposed modification is suitable for hardware implementations of HOG-based detectors, enabling an increase of the detection accuracy or resignation from the use of some hardware-unfriendly operations, such as a spatial Gaussian mask. The results of testing its influence on the brightness changes of test images are also presented. The descriptor may be used in sensor networks equipped with hardware acceleration of image processing to detect humans in the images.

show abstract

A Real-time Scalable Object Detection System using Low-power HOG Accelerator VLSI

Cited by 13 publications

References 9 publications

A 58.6 mW 30 Frames/s Real-Time Programmable Multiobject Detection Accelerator With Deformable Parts Models on Full HD $1920\times 1080$ Videos

A 58.6 mW 30 Frames/s Real-Time Programmable Multiobject Detection Accelerator With Deformable Parts Models on Full HD $1920\times 1080$ Videos

A 58.6mW real-time programmable object detector with multi-scale multi-object support using deformable parts model on 1920×1080 video at 30fps

Histogram of Oriented Gradients with Cell Average Brightness for Human Detection

Contact Info

Product

Resources

About