Tracking Algorithm of Multiple Pedestrians Based on Particle Filters in Video Sequences

Algorithm 1

The entire algorithmic process.

Detection of pedestrians

 For

    (1.1) Perform frame difference and obtain object region;

    (1.2) Build a pedestrian detector using HOG descriptors and SVM;

    (1.3) Detect pedestrians in each frame;

    (1.4) Extract priori knowledge of pedestrians;

 End for

 For

    (1.5) Perform superpixel segmentation according to object region of the last frame;

    (1.6) Obtain confidence map according to (2);

    (1.7) Do random sampling for object confidence map and obtain object;

    (1.8) Update priori knowledge for each object;

 End for

Particle filter tracking

 For object =

    (2.1) Initialize particle state distribution ;

    (2.2) Set initial weight value of feature information ;

 End For

 For

  For target =

    (2.3) Important sampling step

    Propagate and get new particles using (4);

    (2.4) Update the weights

    Compute the observation likelihood function and for each particle using (7);

    Update weight value of features information using (9);

    If

    End if

  End For

 End For

    (2.5) State estimation

    Estimate the object state