Using IMAPCAR, car manufacturers can produce Pre-crash safety systems that can recognize road markers, vehicles and pedestrians in real time, and assist in collision avoidance.

Step 1. The unnecessary elements in the screen image are eliminated.
Step 2. A pedestrian template is taken from the database.
Step 3. The template is superimposed on the image being analyzed.
While the overlay position of the template is gradually moved, the system goes to Step 4, and evaluates the match between the image and the template.
Step 5. If the image conforms to the template of the pedestrian, the system recognizes that the object in the image is indeed a pedestrian.
The system can process more than one image screen at a time; in fact it can simultaneously look at images of the front and the rear of the vehicle to see if they contain pedestrians. In other words, image recognition involves looking at an image, and discerning and identifying the objects (vehicles, pedestrians, white lines, etc.) that you are interested in. By continuously processing a video stream, the system can detect and track changes over time. Parallel processors can process data in batches at high speed, and are very effective in the repetitive processing and verification of images and templates.

IMAPCAR has the following architecture to enable processing at speeds of up to 100 GOPS: 1) 128 processing elements (PE) that carry out SIMD* processing, in which all the elements process the same instruction in unison. 2) Each PE has its own independent memory. By using VLIW** to accelerate processing, IMAPCAR can execute up to four instructions per clock cycle.

*SIMD: Single Instruction, Multiple Data
**VLIW: Very Long Instruction Word