| With the development and improvement of Brain Computer Interface and Hybrid Intelligence Technology, the traditional behavioral experiment technology, which is based on manual observa-tion and statistics, is gradually replaced by a new training mode combining computer vision and brain electrical reward stimulation. The introduction of computer vision has make the analysis of complex behavior patterns of experimental animals possible.In the background of ratbot maze learning, this thesis has studied and designed the ratbot maze experiment system, which can capture rat’s movement in real time and give brain electrical stimulation by the real-time experiment software, and automatically analyze the rat’s performance by the experiment data analysis software. This thesis has mainly done the following.1. Design and deploy the whole maze experiment environment, and introduce the camera cali-bration method used on our CMOS cameras, multi-camera capture and multi-scenario stitch-ing technique.2. Design and implement the real-time experiment software. The real-time experiment software using multithreading asynchronous programming techniques, can record the rat’s behavior, collect behavior data in real time, and give brain electrical stimulation according topredefined status rules.3. Using image algorithms to extract rat’s contour and body centroid. At the same time, be-cause there are many of sheltered dark areas in the experiment environment, and the rat body is flexible and has less features, which makes rat’s movement patterns complex and unpre-dictable. So, this thesis proposed an automatic rat’s head and tail tagging algorithm, which is based on HOG/SVM and the iterative tail-centroid calculation method.4. Design and implement the experiment data analysis software. Assess experiments by analyze the off-line experiment video data, measure the data according to a set of data indicators, such as correct decision rate, total path length, elapse of experiment, different error analysis, etc. |
PDF Full Text Request