In this report, I will illustrate the work of my project on vehicle detection on real time videos using CNNs. The desertation is in french (see desertation.pdf).Soon , ill upload an english version of the summary .
The study concern vehicle detection algorithm called YOLO, knowing that this algorithm is the fastest and precise algorithm on vehicle detection compared to other architectures like regular RNNs and CNNs .
The library used is Tensorflow : https://github.com/tensorflow/tensorflow .
The algorithm used is YOLO and its model DARKNET : https://pjreddie.com/darknet/yolo/.
Link to the notebook : https://drive.google.com/open?id=1alQfjZYJ5BWoZGgHNdGLLCIXFsjkWtVW .
Machine translation is constantly evolving. ML gives the entire world the ability to imporve however is the domain : Buisness or else. it also allows people from diverse countries with distinct languages to stay connected and develop their skills and be up to date with the state of art in AI .
The problem on vehicle detection is : How could we be able to detect vehicles on real time videos (Surveillance cameras) without losing FPS and with a high precision ? How to create a model dedicated for vehicle detection ? or which model should we use ? which algorithm to use ? why did we choose this model ? why using this algorithm and not another one ? is the model choosed/created is perfoming well or not? . The YOLO algorithm is the most famous with his high speed of detection on real time videos especially (not under 30fps) and his detection rate (90% ~ 96%) . The model used is darknet19-53, 53 is a reference to the number of layers contained on the architecture of the model .
The advantage of this algorithm is that he accepts any size of video, and in RGB same as GRAYSCALE. the detection time is on seconds ( ~ 10s ) .
Result 1 :
Result 2 :
