Evaluation and Acceleration of High-Throughput Fixed-Point Object Detection on FPGAs

Abstract

Abstract-The reliance on object or people detection is rapidly growing beyond surveillance to industrial and social applications. The Histogram of Oriented Gradients (HOG), one of the most popular object detection algorithms, achieves high detection accuracy but delivers just under one frame-per-second (fps) on a high-end CPU. FPGA accelerations of this algorithm are limited by the intensive floating-point computations. All current fixedpoint HOG implementations use large bit-width to maintain detection accuracy, or perform poorly at reduced data precision. In this paper we introduce the full-image evaluation methodology to explore the FPGA implementation of HOG using reduced bit-width. This approach lessens the required area resources on the FPGA and increases the clock frequency and hence the throughput per device through increased parallelism. We evaluate the detection accuracy of the fixed-point HOG by applying state-of-the-art computer vision pedestrian detection evaluation metrics and show it performs as well as the original floatingpoint code from OpenCV. We then show our single FPGA implementation achieves a 68.7x higher throughput than a highend CPU, 5.1x higher than a high-end GPU, and 7.8x higher than the same implementation using floating-point on the same FPGA. A power consumption comparison for different platforms shows our fixed-point FPGA implementation uses 130x less power than CPU, and 31x less energy than GPU to process one image.

Low-Power Pedestrian Detection System on FPGA

“…Similarly, the design used some kinds of frame buffer before HOG processing module, which costs memory. The energy consumption of a HOG-based detection system on FPGA is first reported in [2]. In this work, the authors try to reduce the bit-width of the fixed-point representation to boost the performance.…”

“…On the other hand, only extracted features from the image will input the detection algorithm. This approach using HOG (Histogram of Gradients) [1] has proven to have good accuracy in detection [2]. While requiring less memory capacity, it is still a computing-intensive algorithm, which needs a low latency and high-throughput platform.…”

Algorithmic optimisation of histogram intersection kernel support vector machine‐based pedestrian detection using low complexity features

“…Especially its bilinear interpolation method for histogram population and subsequent normalisation take a heavy toll on the processor. Dedicated full hardware implementations of HOG also require complex data flow controllers and arithmetic units [12, 13]. A low complexity variant of the original HOG , HSG‐HIK detector, was proposed in [6] and yields better performance despite being simpler to compute.…”

“…A low complexity variant of the original HOG , HSG‐HIK detector, was proposed in [6] and yields better performance despite being simpler to compute. HSG‐HIK , the predecessor of the work being presented in this paper, uses LUT ‐based HIK‐SVM in place of linear SVM to increase its discrimination power and renders the original HOG‐linear SVM detector, still being employed in dedicated hardware platforms [12, 13], redundant.…”