identifying individual object

[nisarg09]

Hi.. I have implemented the same and getting results as you have shown. The issue here is I need to identify each individual object in each sequence of frame in video(input source). Can you tell me please how to identify/track particular object in consecutive frames from the list of rectangles from given output?

[andrewssobral]

Hi @nisarg09 sorry for my late response.

For object tracking you can use a:

• 1. very basic approach: by a blob tracker algorithm such as cvBlob or OpenCVBlobsLib.
• 2. traditional approach: by a tracker algorithm, please see Introduction to OpenCV Tracker. See also Tracker Algorithms.

For object recognition (e.g. type of the vehicle: car,bus,truck,etc) you will need to build your own classifier. There are several ways to do this, some of them are:

• 1. very basic approach: you use a background subtraction algorithm (e.g. bgslibrary) and classify the objects by its blob size (see simple_vehicle_counting).
• 2. traditional approach: giving the bounding box of the detected vehicles, you extract its features and use a secondary classifier (such as svm) for vehicle type classification.
• 3. more robust approach: use the recent advances on deep convolutional neural networks (Deep CNNs, such as caffe, tiny-dnn, etc) for automatically feature extraction and classification.

However, for both options 2 and 3, you will need to build your own image dataset (with hundreds of images) for training, validation and test.