Reducing impact of industrial wind turbines on wildlife, particularly on raptors, is one of the challenges to carry out for achieving the environmental friendly development of this renewable energy. One of the possible work options is the real time detection of the threatened birds, coupled with an appropriate action (warning signal and/or wind turbine slow down). This is the aim of ProBird, and we present here a feedback compiled from 10 wind farms (42 wind turbines) in France and Germany equipped since more than one year. There are several technical issues in this approach, however the most problematic is to obtain a reliable real time bird detection. We have chosen to handle this part with dedicated high sensitivity IP cameras, monitoring several panoramic views on each wind farms. The video streams of these cameras are grabbed on multicore computer (one CPU core dedicated to one camera, up to 8 cameras managed on a single computer). The raw bird activity is recorded by stacking the pictures generated by all the cameras. This allows to store different position of a bird in the field of view of each camera, during several seconds, in a unique frame. Such a storage induces a drastic decrease of the memory needed to record the raw information out of the camera (20 time less compared to a video file). However, these synthetic long exposure images contain all the information needed for a human to check if there is a bird or not in the monitored area, at a defined time. They also provided a mortality survey by storing pictures of all bird interaction with blades. This row storage is completed by an active detection of bird like objects realized by an algorithm coded in Python. This algorithm is divided in 4 main steps: 1) motion detection with a quick size filter; 2) blades removal based on a first basic shape analysis; 3) enhanced shape analysis to reject clouds and vegetation motion; 4) trajectory (speed, linearity, shape shift) analysis. The comparison between the automated detection managed by this algorithm was conduced on a subsample of 20 000 minutes of record. These 20 000 minutes of record contain 428 detection of bird. For each detection ± 0.024 missed detection are reported while ± 0.16 false detection are induced, mostly by clouds. Influence of meteorological conditions and sun position on these results are discussed In parallel, test of drone detection was managed to define the detection distance of the system on 3 wind farms. The Mavic Pro used is detected at 120 ± 23 m during clear weather, even with complex background (fast moving clouds). Extrapolation of this measurements implies that a reliable detection of a kite can be managed up to 700 m in good conditions. This capacity of detection offers enough time to initiate sequence of acoustic warning and wind turbine stop depending of the behavior of the detected birds. It also open some perspectives to detection and management of small birds like skylarks.