jh's profile - activity

Hi I want to be able to detect cars or people or any object of interest in a live video. The camera will be far away from the target(ie the camera will be in a uav). How many pixels do i need on the target to detect the target ? You dont need to think about it from a uav point of view, the question is independent of that. How many pixels will i need on a car( 200 pixels, 1000 pixels ?), say for the histogram of oriented graidents, or haar classifiers or any object detection method to work reasonably well? If you can provide a reference as well, that will be preferred. Personal experience is fine as well.

I want to use opencv for real time object detection from really high quality dslr video camera feed. How do i select the optimum architecture ? Are there any resources to help in architecture selection? I want something that's powerful, but how do i know if its too powerful increasing my weight, cost and power consumption ? These last 3 variables are very important. I also want standard camera interfaces. For my project my weight needs to be below 2kg, power below 30 W, small form factor and as much processing power tailored to running opencv. Can you a) make a recommendation with a justification and/or more importantly b) provide me with a link that will allow me to quickly choose the best architecture for my unique project? I dont want a 2kg, 30 W system if a beaglebone or some cheap lightweight system will do it at 30g & 2 W. How do i evaluate d architecture's ability to perform my time sensitive application? I want a website that will allow me to search several architectures, specifying criteria/constraints, selecting the best opencv architecture for my needs. Recommendations ?

Thanks, Sorry 4 d nube question