1 / 15

Car detection using a Bayesian network

Car detection using a Bayesian network. Hoiem, Efros, Hebert, Putting objects in perspective , IJCV, 2008. Malagon-Borja, Olac Fuentes Object detection using image reconstruction with PCA , IVC, 2007. Konstantin Speransky. Outline. Problem description

silver
Télécharger la présentation

Car detection using a Bayesian network

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Car detection using a Bayesian network • Hoiem, Efros, Hebert, Putting objects in perspective, IJCV, 2008 • Malagon-Borja, Olac Fuentes • Object detection using image reconstruction with PCA, IVC, 2007 Konstantin Speransky

  2. Outline • Problem description • Local detector based on PCA reconstruction • Bayesian framework • Results • Conclusion and possible improvements

  3. Server Smartphone Parking system design project • Smartphone: • Periodically takes and preprocess pictures • Estimate position based on consecutive GPS measurements • Send this information to the server • Server: • Gather information • Detect cars/free parking spaces from pictures • Match detection results to the map • Answer on requests about presence of free parking lots in a particular region

  4. Local Detector 150x80 Malagon-Borja, Olac Fuentes Object detection using image reconstruction with PCA, IVC, 2007 [2]

  5. Classifier based on PCA reconstruction Reconstruction with car eigenimages Reconstruction with background eigenimages

  6. Similarity metrics Probability of car detection for 3% false detection rate Time in seconds to evaluate 750 windows

  7. Detection pruning 68 -->6 Box 2 Box 1

  8. Motivation to use contextual information • Results from the local detector are often ambiguous • Context plays a crucial role in scene understanding • Car is recognized as a car because: • it’s on the road • it has the right size in comparison to other objects • It’s a chicken and egg problem: • Scene influences detections • Detections influence scene • We need in a unified framework to combine detections and scene understanding • Picture is from [1] Hoiem, Efros, Hebert, Putting objects in perspective, IJCV, 2008 [1]

  9. Scene projection • Pictures are from [1]

  10. Road extraction • Steps: • Select number of reference points • Calculate the Euclidian distance between RGB values of reference pixels and all other pixels • Select only pixels that are relatively close to reference pixels in their RGB values • Find connected region among these pixels that incorporates reference pixels

  11. Bayesian framework Estimation: camera height + horizon Possible objects are acquired from local detector. • Possible values of geometry: • ground • vertical • vertical + ground below

  12. Taking viewpoint into account

  13. Taking viewpoint and geometry into account

  14. More results

  15. Conclusion • Implemented Algorithm: • Bayesian framework combining information from the PCA-based local detector with contextual cues. • Advantages: • High performance local detector, insensitive to blurring • Bayesian network improves results further • Simple ground plane detector • Things to improve: • Computationally intensive local detector • Limitations of the ground plane detector • Objects are independent given camera position

More Related