Search code examples
image-processingmachine-learningartificial-intelligencecomputer-visionobject-detection

Classification of objects from a video ( human, animals, others(cars etc.,) )

Hi I am new to image processing field. My project is to classify the objects in an image/Video. The input is an image/video from surveillance camera. I should classify the objects into three classes. 1) Humans 2) Animals 3) Others (cars). We can consider a fixed background.

Can any one suggest any proven algorithms or papers or opensource code for this purpose? offcourse googling is obvious solution but always an experienced eye is always helpful!

Solution

  • For your purposes, I think the best option is LatSVM because the authors disponibilizes the implementation in MATLAB and the detector models already trained. The already trained models includes:

    • Aeroplane, Boat, Bus, Car, Bicycle
    • Bottle, TV, Sofa
    • Bird, Cat, Horse, Sheep, Dog
    • etc.

    There is also an implementation in OpenCV if you need (found out in this question).

    Other Detector Implementations

    Give a look at the following surveys in Pedestrian Detection:

    1. Pedestrian Detection: An Evaluation of the State of the Art (2012).
    2. Survey of Pedestrian Detection for Advanced Driver Assistance Systems (2010).

    State of Art

    • Pedestrian detection at 100 frames per second. This work proposed by [Benenson et al., 2012] achieves a high detection speed with state-of-art precision [webpage].
    • Object Detection with Discriminatively Trained Part Based Models aka LatSVM Achieved good results on PASCAL 2006, 2007 and 2008. This work doesn't show only for pedestrian detection, but also for cars, cats, horse, sofa etc. You really should give a look at this one [webpage]
    • Human Detection using Partial Least Squares Analysis. [Schwartz et al., 2009] proposed the use of Partial Least Squares analysis which allows to handle high-dimensional space features.

    Classical Works

    The following works has important contributions and are references in every work of pedestrian detection. HOG feature, Integral Image, and Cascade of Rejection are used by several works, including the one by [Benenson et al., 2012].

    • Histograms of Oriented Gradients for Human Detection
    • Fast Human Detection Using a Cascade of Histograms of Oriented Gradients