Research On Multi-features Based Approach For Pedestrian Counting

Counting the number of pedestrians in real time can provide data support forpersonnel scheduling, resource allocation, business strategy, safety control in industry.Vision-based systems have many advantages compared to traditional methods. Theycan be real-time, accurate and interference-free. In recent years, many researchershave made helpful attempts to improve accuracy. However vision-based methodssuffer a lot from illumination changes, shadows or reflections and the crowds.According to this, our method brings in depth features. Depth information helps alot in this area, such as, convenient target extraction, easy for block segmentation,interference free from illumination changes and so on. This paper presents a less costpeople counting system by fusing the depth and IR vision data. Based on vision-basedmethods, an infrared laser is involved in our method. And a traditional camera with aband-pass filter is used to capture the spot image of the region. First, the presentmethod builds the background model based on the spot image of the empty scene.Second, relative depth information is recovered from the extracted changes of spotsbetween this frame and the background. These changed spots make up the foreground.Then, we can get the head region based on the relative depth data. Heads usually turnup as the local maximum value. Now on our method detect the individuals effectively.Finally, a bidirectional matching method is used after the LK Optical Flow algorithmto track the pedestrians and then we carry on people counting. Experiments show that the proposed method can achieve great accuracy in real-time and handle the crowdseffectively.