System and method for partially occluded object detection

Patent metadata

Field	Value
Publication number	US-9785828-B2
Application number	US-201514641506-A
Country	US
Kind code	B2
Filing date	Mar 9, 2015
Priority date	Jun 6, 2014
Publication date	Oct 10, 2017
Grant date	Oct 10, 2017

Abstract

A method for partially occluded object detection includes obtaining a response map for a detection window of an input image, the response map based on a trained model and including a root layer and a parts layer. The method includes determining visibility flags for each root cell of the root layer and each part of the parts layer. The visibility flag is one of visible or occluded. The method includes determining an occlusion penalty for each root cell with a visibility flag of occluded and for each part with a visibility flag of occluded. The occlusion penalty is based on a location of the root cell or the part with respect to the detection window. The method determines a detection score for the detection window based on the visibility flags and the occlusion penalties and generates an estimated visibility map for object detection based on the detection score.

First claim

The invention claimed is:

1. A computer-implemented method for partially occluded object detection, comprising:
obtaining a response map for a detection window of an input image, wherein the response map is based on a trained model and the response map includes a root layer and a parts layer;
determining visibility flags for each root cell of the root layer and each part of the parts layer based on the response map, wherein the visibility flag is one of visible or occluded;
determining an occlusion penalty for each root cell with a visibility flag of occluded and for each part with a visibility flag of occluded, wherein the occlusion penalty is based on a location of the root cell or the part with respect to the detection window, and the occlusion penalty is lower for root cells and parts located in a lower part of the detection window than root cells and parts located in a higher part of the detection window;
determining a detection score for the detection window based on the visibility flags and the occlusion penalties; and
generating an estimated visibility map for object detection based on the detection score.

2. The computer-implemented method of claim 1 , wherein the trained model is a deformable parts model.

3. The computer-implemented method of claim 1 , wherein obtaining the response map comprises determining a cell level response score for each root cell and a parts response score for each part.

4. The computer-implemented method of claim 3 , wherein determining visibility flags for each root cell is based on the cell level response score and a location of the root cell relative to adjacent root cells.

5. The computer-implemented method of claim 3 , wherein determining visibility flags for each part is based on the parts response score and a location of the part relative to overlapping root cells.

6. The computer-implemented method of claim 1 , wherein determining visibility flags comprises applying consistent visibility flags to adjacent root cells.

7. The computer-implemented method of claim 6 , wherein determining visibility flags comprises applying consistent visibility flags to overlapping parts and root cells.

8. A system for partially occluded object detection, comprising:
an image input device receives an input image;
an object detector determines a response map for a detection window of the input image, wherein the object detector determines the response map based on a trained model and the response map includes a root layer and a parts layer;
a processor operatively connected for computer communication to the image input device and the object detector;
a visibility flag module of the processor determines visibility flags for each root cell of the root layer and each part of the parts layer based on the response map, wherein the visibility flag is one of visible or occluded, the visibility flag module of the processor determines an occlusion penalty for each root cell with a visibility flag of occluded and for each part with a visibility flag of occluded, wherein the occlusion penalty is based on a location of the root cell or the part with respect to the detection window, and the occlusion penalty is lower for root cells and parts located in a lower part of the detection window than root cells and parts located in a higher part of the detection window; and
the object detector determines a detection score for the detection window based on the visibility flags and the occlusion penalties and generates an estimated visibility map for object detection based on the detection score.

9. The system of claim 8 , wherein the object detector determines the detection score for the detection window based on a cell level response score of each root cell in the root cell layer and a parts response score based on each part of the parts layer.

10. The system of claim 8 , wherein the visibility flag module of the processor compares a deformable parts model detection score to a predetermined threshold.

11. The system of claim 10 , wherein the visibility flag module of the processor determines visibility flags for the detection window if the deformable parts model detection score meets the predetermined threshold.

12. A computer-implemented method for partially occluded object detection, comprising:
obtaining a response map for a detection window of an input image, wherein the response map is based on a trained model and the response map includes a root layer and a parts layer;
determining visibility flags for each root cell of the root layer and each part of the parts layer based on the response map, wherein the visibility flag is one of visible or occluded;
determining a detection score for the detection window including determining a detection score for each root cell with a visibility flag of visible and each part with a visibility flag of visible and determining an occlusion penalty for each root cell with a visibility flag of occluded and each part with a visibility flag of occluded, wherein the occlusion penalty is based on a location of the root cell or the part with respect to the detection window, and the occlusion penalty is lower for root cells and parts located in a lower part of the detection window than root cells and parts located in a higher part of the detection window; and
generating an estimated visibility map for object detection based the detection score and the occlusion penalty.

13. The computer-implemented method of claim 12 , wherein obtaining the response map comprises determining a cell level response score for each root cell and a parts response score for each part.

14. The computer-implemented method of claim 13 , wherein determining visibility flags for each root cell is based on the cell level response score and a location of the root cell relative to adjacent root cells.

15. The computer-implemented method of claim 14 , wherein determining visibility flags for each part is based on the part response score and a location of the part relative to overlapping root cells.

16. The computer-implemented method of claim 12 , wherein the occlusion penalty is based on a location of the root cell and parts with respect to an image of a pedestrian, wherein the input image includes the image of the pedestrian.

17. The computer implemented method of claim 16 , wherein the occlusion penalty is lower when the root cell or part is located in a lower part of the pedestrian.

Assignees

• Honda Motor Co Ltd

Inventors

• Ayvaci Alper
• Chan Kai-Chi
• Heisele Bernd

Classifications

• G06V10/84

  using probabilistic graphical models from image or video features, e.g. Markov models or Bayesian networks · CPC title

• G06F18/29

  Graphical models, e.g. Bayesian networks · CPC title

• G06V40/10 Primary

  Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands · CPC title

• G06V10/50

  by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis · CPC title

• G06V10/457

  by analysing connectivity, e.g. edge linking, connected component analysis or slices · CPC title