ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 1603.04525
6
94

Pushing the Limits of Deep CNNs for Pedestrian Detection

15 March 2016
Qichang Hu
Peng Wang
Chunhua Shen
Anton van den Hengel
Fatih Porikli
ArXivPDFHTML
Abstract

Compared to other applications in computer vision, convolutional neural networks have under-performed on pedestrian detection. A breakthrough was made very recently by using sophisticated deep CNN models, with a number of hand-crafted features, or explicit occlusion handling mechanism. In this work, we show that by re-using the convolutional feature maps (CFMs) of a deep convolutional neural network (DCNN) model as image features to train an ensemble of boosted decision models, we are able to achieve the best reported accuracy without using specially designed learning algorithms. We empirically identify and disclose important implementation details. We also show that pixel labelling may be simply combined with a detector to boost the detection performance. By adding complementary hand-crafted features such as optical flow, the DCNN based detector can be further improved. We set a new record on the Caltech pedestrian dataset, lowering the log-average miss rate from 11.7%11.7\%11.7% to 8.9%8.9\%8.9%, a relative improvement of 24%24\%24%. We also achieve a comparable result to the state-of-the-art approaches on the KITTI dataset.

View on arXiv
Comments on this paper