Multi-people detection and tracking is a basic function for service robots when they serve people and interact with people. However, there exists a big challenge because of complex environment and occlusion of multiple people. To detect people, a new method based nearest neighbor algorithm combiningleg detection of laser and body detection of Kinect is proposed and it improves the accuracy and completeness of detection. For multi-people tracking, an improved particle filter is proposed to predict the people's position and velocity, which can overcome the problem of heavy computational amount and particle degradation. In the experimental stage, a turtlebot robot with laser and Kinect is used to test the method in a real indoor environment. Results show the method can detect and track people in real time, and the method is robust and accurate.