System and method for compressing video data

A non-transitory computer readable medium comprising computer executable instructions for:

• configuring an object of interest detector to detect whether objects of interest are located in the video data according to one or more classifiers c;
• training the one or more classifiers c to identify the objects of interest for a specific application using data collected from a plurality of video capture devices in a video processing system by applying one or more machine learning methods;
• configuring an encoder to transform an image region I using a non-linear transformation Φ(I) into an ordered set of basis transformation parameters {right arrow over (θ)}, where {right arrow over (θ)}=Φ(I) and I≈Φ−1({right arrow over (θ)});
• collecting video data at a video capture device;
• applying the non-linear transformation Φ(I) to the image region I from the video data to generate compressed video data comprising the set of basis transformation parameters {right arrow over (θ)}associated with the objects of interest;
• and
• sending the compressed video data from the video capture device to the video processing system to enable the video processing system to decode and reconstruct the video using Φc−1({right arrow over (θi )}), for each object of interest i associated with the corresponding object classifier c.

A video compression module comprising at least one processor and at least one memory, the at least one memory comprising computer executable instructions for:

• configuring an object of interest detector to detect whether objects of interest are located in the video data according to one or more classifiers c;
• training the one or more classifiers c to identify the objects of interest for a specific application using data collected from a plurality of video capture devices in a video processing system by applying one or more machine learning methods;
• configuring an encoder to transform an image region I using a non-linear transformation Φ(I) into an ordered set of basis transformation parameters {right arrow over (θ)}, where {right arrow over (θ)}=Φ(I) and I≈Φ−1({right arrow over (θ)});
• collecting video data at a video capture device;
• applying the non-linear transformation Φ(I) to the image region I from the video data to generate compressed video data comprising the set of basis transformation parameters {right arrow over (θ)}associated with the objects of interest;
• and
• sending the compressed video data from the video capture device to the video processing system to enable the video processing system to decode and reconstruct the video using Φc−1({right arrow over (θi )}), for each object of interest i associated with the corresponding object classifier c.