Adaptive Boosting and Cascade Classifier Based Human Detection and Tracking Using Laser Range Sensor

This paper presents the development of a system for human detection using LiDAR sensor for simulation and real implementation. The scan data from LiDAR is then converted into bird's eye images. A shape representing human feet is labeled as a human and other objects are labeled as not a human. Then, by using a classifier method, it will be determined whether the detected is a human foot or not. If the human foot is detected then the data will be used to predict the position of humans. Based on the detected position, then the calculation process to obtain the relative position of humans to the sensor location is carried out. The system is evaluated using a simulation and real implementation experiments. We compared 3 different methods, namely HAAR, HOG, and LBP that are implemented using adaptive boosting and cascade classifier. Testing accuracy of HAAR method achieved 100% for simulation and 95% for real, HOG achieved 70% for simulation and 60% for real, and finally, LBP achieved 86.7% for simulation and 85% for real implementation.