Real Time Object Detection

Real Time Object Detection using optical sensors represents one of the most challenging problems in pattern recognition. Objects can vary in shape, size, color and their appearance will depend on their pose. Detection algorithms also face the problem of objects with a huge within class variability. Finally, complex outdoor environments, cluttered background, etc, difficult seriously object extraction.

Real Time Algorithm

Most of the real time detection algorithms follow the two-step object detection strategy: hypothesis generation, identifying the possible location of objects candidates, and hypothesis verification, validating these candidates. Following this idea, the Cascade of Adaboost Classifiers, proposed by Viola & Jones, is widely used nowadays. First layers (G1 and G2), composed of simple classifiers, generate regions candidates eliminating most of the background. Final layers (G3), composed of complex classifiers, eliminate all non-object regions obtaining the object position.

The solution reduces the information in each frame to the location and spatial configuration of each hot spot present in the frame. The proposed method successfully segments the image with a total processing delay equal to the acquisition time of one pixel (that is, at the video rate). This processing delay is independent of the image size. The solution is not tied up to one specific camera, and may be used with several infrared cameras with minor adjustments. FPGA area equations are also presented in order to calculate the needed FPGA size for a particular application.


Classifier Cascade

Detection and Classification

After the object is identified in the image, it can also be classified in object sub-classes. For face detection, the algorithm must identify facial expressions like smile, surprise, anger, disgust, etc. In vehicle detection, it can be useful to know if the nearby vehicle is a car or a truck. Applications using person detection must detect a being who is possible waiting, walking, running, fighting, etc.


Real Time Face detection


Real Time Pedestrian Detection


Real Time Vehicle Detection

Viola, P., Jones, M.: Robust real-time object detection. International Journal of Computer Vision (2002)