RESEARCH

Image & Vision

Investigates Image data fusion techniques that combine image and track data from multiple sensors to achieve improved accuracies and more specific inferences than could be achieved by the use of a single sensor alone. Our aim is to explore the state-of-the-art image processing algorithms for achieving effective data fusion as in:

Tracking

작성자 관리자 날짜 2021-04-08 14:51:27 조회수 31

1. Region based tracking        

 The idea here is to identify a connected region in the image - a "blob"-associated with each vehicle and then track it over time using a cross-correlation measure. This approach works fairs well in free-flowing traffic. However, under congested traffic conditions, vehicles partially, which makes the task of segmenting individual vehicles difficult. Such vehicles will become grouped together as one large blob in the foreground image.

2. Feature based tracking      

 Another approach to tracking abandons the idea of tracking objects as a whole but instead tracks sub-features such as distinguishable points or lines on the object. The advantage of this approach is that even in the presence of partial occlusion, some of the sub-features of the moving object remain visible. The technology of tracking points and line features in Tracking filtering formalism(Kalman, EK, IMM, IMMPDA etc) is well developed in the computer vision community. Since a vehicle could have multiple sub-features, the new problem then is that of grouping - what set of features belong to the same object.

3. Lip tracking       

 This demo is used "Active Shape Models"(ASM) algorithm. The ASM is useful to recognize a specific object. Here we want to track lip's shape as the object. First we made the Point Distribution Model about open mouth. The Open mouth's model consist of 30 points. Each point is moved toward the position with a high value of gradient.

4. Human and Vehicle Tracking using Skeleton     

 

  • Set the initial joint model of human and vehicle
     
  • Tracking the skeleton using the matching of joint point of previous model and current model
     
  • Tracking Step

              - Finding the object region by updating the background using adaptive background estimation
              - Set the Joint point of detected region
              - Make the filterbank using joint points
               - Make the matching point through the similarity with detected region in frame
               - Construct the skeleton using matching point

 

5. Active Appearance Model (AAM)          

  • Linear modeling of feature vectors
     
  • Shape and texture of object can be modeled by appearance parameter only.
     
  • AAM is very useful approach for tracking of deformable objects.

댓글 (0)

등록된 댓글이 없습니다.
작성 권한이 없습니다.