Safety system for human detection in robot workspace based on AdaBoost algorithm and support vector machine

U ovom diplomskom radu, predstavljen je sustav za upravljanje robotskim manipulatorom temeljen na prepoznavanju i praćenju ljudi. Rad je nastao kao dio projekta HDR (engl. Humans Detected by Robots) u suradnji s tvrtkom Danieli Systec d.o.o. Korišteni sustav za prepoznavanje i praćenje ljudi temelji se na AdaBoost algoritmu i stroju s potpornim vektorima, koristeći HOG deskriptore i Haar značajke. Robotski sustav sastoji se od robotskog manipulatora ABB-IRB 2400L, RGB-D kamere Microsoft Kinect v1 te ROS razvojnog okvira. Cjelokupni sustav implementiran je i ispitan na ROS Kinetic i ROS Melodic distribucijama. Eksperimentalna evaluacija provedena je na HDR skupu podataka, pri čemu su računati pokazatelji kakvoće odziv, preciznost i točnost. Provedena evaluacija odnosi se na testiranje detektora, predviđanja brzina i zaustavljanja robota. Najveća postignuta vrijednost odziva iznosi 87.5%, dok najveće vrijednosti preciznosti i točnosti iznose 100%. Rezultati variraju u ovisnosti o situaciji i vrsti testa.In this thesis, a system for controlling robotic manipulators based on human detection and tracking is presented. The thesis was written as a part of the HDR (Humans Detected by Robots) project in cooperation with Danieli Systec. The system used for human detection and tracking is based on the AdaBoost algorithm and Support Vector Machines (SVM), using HOG descriptors and Haar features. The robotic system consists of an ABB-IRB 2400L robotic manipulator, a Microsoft Kinect v1 RGB-D camera, and ROS development framework. The entire system was implemented and tested on ROS Kinetic and ROS Melodic distributions. Experimental evaluation was performed on the HDR dataset with recall, precision, and accuracy used as metrics. The evaluation includes detector tests, speed prediction tests, and robot stopping tests. The highest achieved value of recall was 87.5%, while the highest value of precision and accuracy was 100%. The results vary depending on the situation and the type of test